Applied Water Science

AI-Driven Predictive Models for Mapping and Mitigating Endocrine Disrupting

Chemicals in Water Systems
--Manuscript Draft--

Manuscript Number: AWSC-D-25-00134

Full Title: AI-Driven Predictive Models for Mapping and Mitigating Endocrine Disrupting
Chemicals in Water Systems

Article Type: Review Article

Corresponding Author: Malik Tabarak, PhD

Jimma University
ETHIOPIA

Corresponding Author Secondary

Information:

Corresponding Author's Institution: Jimma University

Corresponding Author's Secondary

Institution:

First Author: A S Vickram

First Author Secondary Information:

Order of Authors: A S Vickram

Shofia Saghya Infant

A Saravanan

G Gulothungan

Gulothungan G

Hitesh Chopra

Tabarak Malik, PhD

Order of Authors Secondary Information:

Funding Information:

Abstract: Endocrine disruptors, such as bisphenol A, phthalate esters, and synthetic estrogens,
can accumulate and persist in the environment, causing serious threats to human
health and the ecology of aquatic systems. Conventional detection and mitigation
approaches are typically time-consuming, and not precise to determine hotspots of
contamination. This study investigates the application of AI (Artificial Intelligence)-
technology-based predictive models to facilitate the mapping and find solutions for
reducing Endocrine-disrupting chemicals (EDCs) in water systems. Various machine
learning algorithms (including random forests, support vector machines, deep neural
networks, etc.) can analyse the large volume of data related to the concentrations of
EDCs, their physicochemical properties, and distribution patterns in geographical
areas. The qualitative results portray that models can predict the occurrence of EDC in
water with 92% accuracy by analysing water quality parameters like pH, turbidity, and
temperature. However, in pilot studies, quantitative studies show that optimization of
AI-assisted mitigation strategies, such as advanced oxidation processes (AOPs), and
also the adsorption systems reduce endocrine-disrupting chemicals concentrations by
over 85%. For example, in a sample of wastewater, the use of AI-optimized activated
carbon filters decreased bisphenol A concentration from 18 µg/L to 120 µg/L.
Examples from urban water bodies demonstrate how the model can predict high-risk
areas and how to focus intervention efforts. These results highlight the ability of AI to
be a disruptive technology for the monitoring of water quality and the health of society.

Suggested Reviewers:

Opposed Reviewers:

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Title Page including all authors names and contact details

AI-Driven Predictive Models for Mapping and Mitigating Endocrine Disrupting

Chemicals in Water Systems

Vickram A S1, Shofia Saghya Infant1, Saravanan A1, Bhavani Sowndharya B1, Gulothungan G2,
Hitesh Chopra3, Tabarak Malik4

1
Department of Biotechnology, Saveetha School of Engineering, SIMATS, Chennai, 602105,
India

2
Department of Electronics and Communication Engineering, SRM Institute of Science and
Technology, SRM Nagar, Kattankulathur-603203, Chengalpattu District, Tamil Nadu.

3
Centre for Research Impact & Outcome, Chitkara College of Pharmacy, Chitkara University,
Rajpura, 140401, Punjab, India

4
Department of Biomedical Sciences, Institute of Health, Jimma University, Jimma, 378,
Ethiopia.

*Corresponding Authors: [Link]@[Link]; [Link]@[Link]

Cover Letter

Cover Letter

Date/ 30/01/2025

Dear Editor,
We would like to submit the entitled manuscript “AI-Driven Predictive Models for Mapping
and Mitigating Endocrine Disrupting Chemicals in Water Systems” for the consideration
of publication in the Journal Applied Water Science. The manuscript has been approved by all
authors and has never been published, or under the consideration for publication elsewhere.

We declare the following:

1. The article is original.
2. The article has been written by the stated authors who are ALL aware of its content and
approve its submission.
[Link] article has not been published previously
4. the article is not under consideration for publication elsewhere
5. No conflict of interest exists.
6. No human subjects are involved
7. If accepted, the article will not be published elsewhere in the same form, in any language,
without the written consent of the publisher.

Yours Sincerely,
Blinded Manuscript, excluding author names & affiliation Click here to view linked References

AI-Driven Predictive Models for Mapping and Mitigating Endocrine Disrupting

1
2 Chemicals in Water Systems
3
4
5
6
7
8
9
10
11
12 Abstract:
13
14
15
Endocrine disruptors, such as bisphenol A, phthalate esters, and synthetic estrogens, can
16 accumulate and persist in the environment, causing serious threats to human health and the
17
18 ecology of aquatic systems. Conventional detection and mitigation approaches are typically
19
20 time-consuming, and not precise to determine hotspots of contamination. This study
21
22 investigates the application of AI (Artificial Intelligence)-technology-based predictive models
23
24 to facilitate the mapping and find solutions for reducing Endocrine-disrupting chemicals
25
26
(EDCs) in water systems. Various machine learning algorithms (including random forests,
27 support vector machines, deep neural networks, etc.) can analyse the large volume of data
28
29 related to the concentrations of EDCs, their physicochemical properties, and distribution
30
31 patterns in geographical areas. The qualitative results portray that models can predict the
32
33 occurrence of EDC in water with 92% accuracy by analysing water quality parameters like
34
35 pH, turbidity, and temperature. However, in pilot studies, quantitative studies show that
36
37
optimization of AI-assisted mitigation strategies, such as advanced oxidation processes
38 (AOPs), and also the adsorption systems reduce endocrine-disrupting chemicals
39
40 concentrations by over 85%. For example, in a sample of wastewater, the use of AI-optimized
41
42 activated carbon filters decreased bisphenol A concentration from 18 µg/L to 120 µg/L.
43
44 Examples from urban water bodies demonstrate how the model can predict high-risk areas and
45
46 how to focus intervention efforts. These results highlight the ability of AI to be a disruptive
47 technology for the monitoring of water quality and the health of society.
48
49
50 Keywords: Endocrine-disrupting chemicals (EDCs), Artificial intelligence (AI), Predictive
51
52 models, Water quality, Machine learning, Advanced oxidation processes (AOPs), Adsorption
53
54 systems, Bisphenol A, Phthalates
55
56
57
58
59 1. Introduction
60
61
62 1
63
64
65
 Endocrine disrupting chemicals (EDCs) are substances that disrupt hormonal
1
2 functioning in humans and animals, posing significant health hazards. They interfere with the
3
4 reproductive, immunological, and metabolic systems by mimicking, blocking, or modifying
5
hormone function. Water systems are the most common route of exposure to EDCs
6
7 (Gonsioroski et al., 2020). Industrial, agricultural, and residential activities pollute EDCs
8
9 through wastewater, runoff, and landfill leaching (OECD 2023). Because they are persistent,
10
11 EDCs accumulate in aquatic habitats, bioaccumulate in species, and biomagnify up the food
12
13 chain, endangering both wildlife and human health (Sharma et al. 2020). Aquatic life,
14
15 particularly fish and amphibians, is very susceptible to EDC-induced hormonal changes. The
16 implications include reproductive difficulties such as decreased fertility, skewed sex ratios, and
17
18 intersex abnormalities. Developmental abnormalities and decreased immunity endanger
19
20 biodiversity by disrupting predator-prey relationships and the ecological balance (Corpuz-
21
22 Hilsabeck and Culty 2023). The persistence of EDC pollution can contribute to population loss
23
24 and environmental instability (Kessler et al. 2019). Human exposure to EDCs occurs through
25
26 drinking water, recreational activities, and the consumption of contaminated seafood (Jones et
27 al. 2020). The health effects vary depending on the chemical, exposure level, and sensitivity of
28
29 the individual. EDCs induce reproductive abnormalities, affect sexual development, and
30
31 increase the risk of hormone-related cancers such as breast, prostate, and ovarian cancer
32
33 (Rehman et al. 2018; Vilela et al. 2018). Developmental exposure has long-term effects on
34
35 growth, neurodevelopment, and immunological function. Furthermore, EDCs are associated
36
37
with metabolic disorders, obesity, diabetes, and cardiovascular disease (Gaston et al. 2020).
38 Reducing EDC contamination necessitates an integrated strategy. Advanced wastewater
39
40 treatment technologies must be developed to improve EDC elimination. Regulations should
41
42 limit or prohibit dangerous EDCs. Water systems should be continuously monitored to
43
44 determine pollution levels and trends. Public awareness campaigns can help educate people
45
46 about EDC risks and water safety. Further study is needed to better understand EDC sources,
47
48
environmental fate, and cleanup options (Pironti et al. 2021). Addressing the problem of EDCs
49 in water systems is critical for both the environment and human health. A comprehensive
50
51 strategy that includes regulation, technical innovation, surveillance, and education will assist
52
53 to alleviate the long-term impacts.
54
55
56
As artificial intelligence transforms numerous industries, it also improves the efficiency,
57 precision, and scalability of environmental monitoring. AI and its various applications only
58
59 allow for a deeper understanding of environmental systems and the identification of new
60
61
62 2
63
64
65
 dangers in order to develop effective protection methods. Data analysis is one such application.
1
2 This is a methodology in which AI processes massive datasets from sensors, satellites, and
3
4 citizen science initiatives considerably faster and more accurately than traditional methods
5
(Maganathan et al. 2020; Hino et al. 2018). This technology allows academics and
6
7 environmental managers to trace pollution sources, monitor pollutant transport, and evaluate
8
9 the efficacy of cleanup methods. Predictive modeling is another important role for AI. AI can
10
11 anticipate air and water quality, climate change effects, and species distribution using historical
12
13 environmental data, allowing for more informed decision-making (Rozhok et al. 2021;
14
15 McGovern et al. 2022). Furthermore, AI-powered automation enhances environmental
16 monitoring by deploying robots and drones to collect data from remote or hazardous sites,
17
18 decreasing manual labor and human risk (McClure et al. 2020; Zhou 2023). AI can also increase
19
20 data dependability by improving quality control procedures. Smart environmental monitoring
21
22 systems have improved by incorporating AI into sensor networks to provide continuous real-
23
24 time monitoring and early warning of environmental dangers. Such systems help to monitor
25
26 pollution-related incidents, wildlife population trends, and optimize natural resource
27 management (Ullo and Sinha 2020). However, the remaining hurdles include high-quality
28
29 training data, AI model interpretability, and maintaining forecast transparency (McGovern et
30
31 al. 2022). Other ethical problems, such as data privacy and algorithmic bias, must be solved
32
33 before AI can be used responsibly for environmental monitoring (OECD 2022; Burti et al.
34
35 2024). Table 1 explains the AI-based methods for monitoring endocrine disrupting chemicals
36
37
in water systems
38
39
40
41
42
43
44 Table 1: AI-Based Methods for Monitoring Endocrine Disrupting Chemicals in Water Systems
45
46 (Metcalfe et al. 2022; Kavianpour et al. 2024).
47
48
49 Dataset Input Output Pollutants
50 Water quality Physicochemical Classification of Bisphenol A,
51
52 monitoring data properties of water EDC presence Phthalates
53
54 Historical water Time-series pollutant Predicted EDC Pesticides,
55
56 quality data data concentration Pharmaceuticals
57
58 levels
59
60
61
62 3
63
64
65
 Geospatial and Geographic and flow Likely sources of Industrial chemicals,
1
2 hydrological data data EDCs Agricultural runoff
3
4 Environmental Images of affected Risk level Hormones, Heavy
5
6
impact studies habitats predictions metals
7 IoT sensor streams Sensor signal data Alerts for EDC BPA, Parabens,
8
9 threshold breaches Phenols
10
11 Treatment plant Pre- and post- Efficiency metrics Microplastics, Steroids
12
13 performance data treatment water for EDC removal
14
15 samples
16
17 Hydrodynamic and Flow dynamics and Dispersion patterns PCBs, Dioxins
18
19
EDC data EDC properties in aquatic systems
20 Multi-decade Historical pollutant Projected EDC Pharmaceuticals,
21
22 monitoring data levels concentrations Pesticides
23
24 Treatment plant Operational Optimal process Endocrine disruptors,
25
26 configurations parameters configurations Chlorinated
27
28 compounds
29
30 Untargeted Mass spectrometry Grouping of Unknown emerging
31 chemical analysis data unknown chemicals EDCs
32
33 data
34
35
36
37
38 2. Characterization of Endocrine-Disrupting Chemicals
39
40 On the other side, some natural and artificial compounds known as endocrine-
41
42 disrupting compounds cause negative consequences in humans and wildlife since their
43
44 interference with the normal operation of the endocrine system leads to various effects. These
45
46 compounds produce health effects including developmental anomalies, reproductive problems,
47
48
and metabolic disorders (Sharma et al. 2020). Endocrine disruptors can offer their mechanism
49 by means of imitating natural hormones, blocking hormone receptors, disrupting synthesis or
50
51 metabolism, or a combination of these (Gaston et al. 2020). Some endocrine disruptors prevent
52
53 natural hormones from acting by binding to hormone receptors; others block these receptors
54
55 and therefore imitate the effects of natural hormones. EDCs could also affect the production or
56
57 breakdown of the hormone and therefore disturb the usual endocrine activity (Papalou et al.
58
2019).
59
60
61
62 4
63
64
65
 2.1 Sources and Types of EDCs
1
2
3
Endocrine-disrupting chemicals exist in many natural and artificial forms affecting the
4 endocrine system to alter its balance and therefore create potential negative health
5
6 consequences on people and wildlife. EDCs enter the environment through many different
7
8 routes. Some are released through industrial or municipal wastewater discharges, thus entering
9
10 water bodies via industrial discharges or wastewater treatment plants (Tang et al. 2020).
11
12 Agricultural runoff containing pesticide and herbicide and other agricultural chemicals can also
13
14
lead to water source contamination. EDCs enter groundwater and surface water by leaching
15 from landfills and contaminated sites. Finally, naturally occurring substances such as plant
16
17 hormones also can cause endocrine disruption. These broadly can be classified into several
18
19 groups (Mishra et al. 2023). Phthalates, common plasticizers used to impart flexibility to
20
21 plastics, are found in a wide variety of consumer applications such as food packaging, toys,
22
23 and cosmetics. Their relentless use and capability to seep into the surroundings throughout
24
manufacturing, use, and disposal raise worries about water pollution and possible human health
25
26 hazards (Rehman et al. 2018). Phthalates could be exposed to humans by several routes:
27
28 ingestion, inhalation, and dermal contact. Studies has linked phthalate exposure with metabolic
29
30 disorders, developmental and reproductive difficulties, and certain cancers. (Gaston et al.
31
32 2020).
33
34 Bisphenol A is another commercially major substance commonly found in
35
36 polycarbonate plastics, epoxy resins, and thermal paper receipts. BPA might leach into the
37
38 environment like phthalates and therefore present a human health hazard. Mimicking estrogen
39
40 and interfering with typical hormonal activity, BPA is an endocrine disruptor which links BPA
41
42 with developmental issues, metabolic disorders, and fertility problems (Gonsioroski et al. 2020;
43 Jones et al. 2020).Like those found in hormone replacement therapy and birth control pills,
44
45 synthetic estrogens get into water systems via wastewater discharge. Very low amounts of these
46
47 strong substances can disturb the endocrine systems of fish life, causing developmental issues
48
49 and infertility. The existence of synthetic estrogens in the water bodies presents serious dangers
50
51 concerning their possible influence on aquatic systems and the likelihood of people being
52
53 exposed through drinking water (Almazrouei et al. 2023; Grzegorzek et al. 2024). Most
54 agricultural chemicals—for instance, pesticides and herbicides—can have endocrine-
55
56 disrupting qualities. Running off farms, they can easily be deposited into rivers and lakes,
57
58 where they might damage aquatic species and serve as possible sources of drinking water
59
60 pollution. These toxins cause developmental disorders, infertility, and other negative health
61
62 5
63
64
65
 effects by disrupting the endocrine systems of water organisms (Mitra et al. 2024; Abaineh et
1
2 al. 2024). Persistent organic pollutants, including PCBs, dioxins, and per- and polyfluoroalkyl
3
4 compounds, are major EDCs because they remain in the environment and bioaccumulate along
5
food chains, posing risks to human and environmental health. Their exposure is linked to
6
7 developmental abnormalities, immunotoxicity, cancer, and reproductive issues (Mandal et al.
8
9 2024; Li et al. 2024). Natural hormones like estrogens and androgens can also be EDCs if they
10
11 reach water systems via runoff or sewage, affecting marine life (Vilela et al. 2018).
12
13 Pharmaceuticals in water bodies endanger ecosystems and human health (Gonsioroski et al.
14
15 2020; Thacharodi et al. 2023).
16
17 2.2 Physicochemical Properties Affecting EDC Behaviour in Aquatic Environments
18
19
20 The physicochemical properties primarily govern the behavior of endocrine-disrupting
21
22
chemicals in aquatic environments, and these properties therefore significantly influence
23 EDCs' transport pathways, ultimate environmental fate, and bioavailability to aquatic
24
25 organisms. It has been shown that, among other properties, the simultaneous solubility of EDCs
26
27 in water directly impacts the distribution and transport in aquatic systems (Ahn and Jeung
28
29 2023). Highly soluble EDCs readily dissolve in water and can be transported over long
30
31 distances, therefore affecting wide areas. In contrast, poorly soluble EDCs tend to bind to
32
sediments or suspended particles, limiting their transport but possibly generating localized
33
34 hotspots of contamination. Hydrophobic EDCs bioaccumulate since they are preferentially
35
36 partitioned into the fatty tissues of aquatic organisms owing to an incapacitating aversion for
37
38 water (Gałązka and Jankiewicz 2022). The bioaccumulation process, therefore, could increase
39
40 the concentrations of EDCs that bio magnify along the food chain. Lipophilic EDCs easily
41
42 penetrate cell membranes and possibly bind to cell receptors, thereby interfering with endocrine
43 function. Persistent EDCs resist environmental degradation and may continue to expose aquatic
44
45 organisms for a long time or get accumulated in them, answering challenges in remediation
46
47 efforts since the chemicals get retained in the environment for a long time (Azizi et al. 2022).
48
49 Volatile EDCs could evaporate from the waterbody and become airborne for long-range
50
51 transportation and hence, a source of air contamination in areas far removed from the original
52
53 source. In such instances, atmospheric transport could be contributing toward the global
54 distribution of specific EDCs. Partitioning coefficient describes the distribution of EDCs
55
56 among various environmental compartments like water, sediment, and air, thereby playing an
57
58 important role in determining EDC bioavailability and potential biomagnification in aquatic
59
60 organisms (Kumar et al. 2020). For example, a high partitioning coefficient toward sediment
61
62 6
63
64
65
 indicates that an EDC is more likely to accumulate in sediments, which lowers the
1
2 bioavailability of the EDC in the overlying water. Therefore, understanding these
3
4 physicochemical properties is critical in predicting the behavior of EDCs in aquatic
5
environments, assessing potential risks to aquatic life and human health, and strategizing on
6
7 their effective planning, management, and mitigation (Stiefel and Stintzing 2023). Advanced
8
9 analytical techniques—for example, high-performance liquid chromatography linked to MS—
10
11 are used to detect and quantify EDCs from environmental samples. Meanwhile, the
12
13 development of computational modeling to predict the fate and transport of EDCs in aquatic
14
15 systems will also serve as supportive information for risk assessment and management (Hino
16 et al. 2018; Jones et al. 2020). Figure 1 presents a diagram illustrating various sources and
17
18 types of environmental endocrine-disrupting chemicals (EDCs).
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48 Figure 1: Sources and Categories of Environmental Endocrine-Disrupting Chemicals (EDCs)
49
50
51
52
53 3. Artificial Intelligence in Water Quality Monitoring
54
55 Water quality monitoring presents a major new opportunity for change, and artificial
56
57 intelligence gives us the opportunity to significantly improve pollution detection, prediction,
58
59 and management, especially where the pollution issue is threadlike with complex
60
61
62 7
63
64
65
 environmental occurrences such as endocrine-disrupting chemicals. Traditional classical
1
2 methods involved so much manual sampling, laboratory analysis, and often a prolonged time
3
4 frame to reach results, making it expensive and limited in its monitoring ability (Ullo and Sinha
5
2020). Artificial intelligence applications could integrate various data streams from sensor
6
7 networks, remote sensing apps, and historical water quality records. This particular ability for
8
9 synthesizing and evaluating information from several sources allows water quality evaluation
10
11 be more thorough and in real time (Rozhok et al. 2021). ). In water quality data, artificial
12
13 intelligence algorithms can detect patterns and draw connections that might otherwise be
14
15 unknown to standard approaches. Since machine learning algorithms trained to detect small
16 changes in water quality indicators might therefore recognize even minor fluctuations in water
17
18 quality, suggesting the existence of EDCs at very low levels, early alert would help to stop the
19
20 more widespread spread of pollution and therefore lower the potential health risks from EPC
21
22 exposure (McClure et al. 2020; Auyb et al. 2024).
23
24
3.1 Machine Learning Algorithms for Predictive Modelling
25
26
27 Regarding water quality management, machine learning is a significant predictive
28
29 modelling tool. Including endocrine-disrupting chemical levels, the algorithms can use
30
31 historical water quality data to create models for forecasting future EDCs in water. In order to
32
offer more precise and complete forecasts, these models might as well employ different sets of
33
34 parameters including environmental conditions (temperature, rainfall), land use patterns
35
36 (agriculture, industrial discharge), and data on wastewater discharge (Agrawal et al. 2021). By
37
38 means of predictive modeling also one could anticipate likely events of EDC contamination
39
40 and proactively act to curb it. The process enables officials to take focused measures including
41
42 more monitoring, public awareness, or short water use limitations by pinpointing locations and
43 times with great chances of pollution. For treatment plants to adjust their procedures as
44
45 required, they have to forecast EDC levels in drinking water sources (Hasan 2021). For
46
47 example, if an increased EDC concentration is expected, the treatment plant would need to
48
49 install better filtration or else apply corrective therapy to guarantee safe drinking water
50
51 availability. Applying this forward-thinking plan can help to lower EDC exposure and defend
52
53 the general public's health. By discerning regions particularly sensitive to EDC pollution,
54 therefore setting a priority region for monitoring and remediation efforts, ML algorithms might
55
56 be employed (Li et al. 2021). This resource optimization enables governments to better decide
57
58 when and where to interfere so that their interventions have greatest effect and the overall cost
59
60 of water quality management is kept to a bare minimum. This targeted strategy can also help
61
62 8
63
64
65
 to identify sources of EDC pollution and support the application of remedies to stop any future
1
2 contamination by means localizing. Integrating of ML models with sensor data in real-time
3
4 allows for perpetual water quality monitoring and dynamic management (Zhong et al. 2021).
5
These models can update their predictions and assessments regarding water quality conditions
6
7 on the basis of incoming data for more precise assessments. This adaptive management allows
8
9 for a dynamic and responsive approach to managing water quality that allows the authorities
10
11 to change their management strategies as various conditions arise (Ahmed et al. 2020).
12
13 Compared to traditional approaches, other advantages of ML-based models for predicting
14
15 water quality include accuracy and productivity. These algorithms operate on huge datasets to
16 the end of recognizing patterns and relationships complicated for human analysis. Hence the
17
18 foregoing accuracy and productivity contribute to enhanced water quality management systems
19
20 and more reliable protection toward public health and the environment (Hmoud Al-Adhaileh
21
22 and Waselallah Alsaade 2021).
23
24
3.2 Role of Big Data in AI-Driven Detection of Endocrine Disrupting Chemicals
25
26
27 Big data analytics is critical for identifying EDCs using AI. The availability of vast
28
29 datasets from many sources enables the training and validation of AI models. EDC
30
31 contamination is determined using historical and real-time EDC concentration data in water
32
bodies, water quality indicators (pH, dissolved oxygen, turbidity, nutrients), and land use
33
34 patterns (agricultural, industrialization, urbanization). Temperature, rainfall, and hydrological
35
36 flow are examples of environmental conditions that influence EDC transport and fate.
37
38 Demographic and socioeconomic factors determining EDC exposure and susceptibility are also
39
40 important inputs. AI finds complicated patterns in vast datasets that would be difficult to detect
41
42 using traditional statistical methods (McClure et al. 2020). It detects minor fluctuations in
43 EDCs in water quality that normal approaches usually overlook, and forecasts EDC incidence
44
45 using historical and environmental data. AI can identify pollution sources by studying land use
46
47 and industrial discharge (Auyb et al., 2024). Including AI-driven data on EDC concentrations,
48
49 water quality, and ecological/human health impacts improves risk assessment. This
50
51 comprehensive approach improves understanding of EDC interactions with environmental
52
53 variables and human behaviors, resulting in more effective risk reduction methods (Zulkifli et
54 al. 2018). Different large data sources aid in EDC detection. The sources include typical water
55
56 monitoring stations, which provide continuous records on EDC levels. Advanced IoT sensor
57
58 networks enable real-time, high-resolution water quality monitoring. Remote sensing via
59
60 satellites and aerial pictures collects data on land use and plant health, which are predictors of
61
62 9
63
64
65
 EDC concentrations (Hassan and Woo 2021; Shah et al. 2023). Citizen science entails
1
2 organizing volunteers to contribute to observations of water quality, whereas social media
3
4 provides insights into real-time contamination incidents and public concern. Governmental,
5
institutional, and nongovernmental repositories include further relevant information about
6
7 EDCs (Mix et al. 2020). This wide approach improves monitoring, assessment, and
8
9 intervention for EDC exposure.
10
11
12 3.3 Applications of AI in Environmental Systems
13
14 Artificial intelligence in environmental systems goes farther than just water quality
15
16 monitoring to encompass every aspect of environmental control. Enhancing efficacy and
17
18 resource use, AI helps to optimize the water treatment procedure. AI algorithms might, for
19
20 instance, analyze current information from water treatment facilities for setting treatment
21
22
parameters, maximizing chemical dosages, and lowering energy use (Li et al. 2021). Using
23 artificial intelligence to optimize treatment operations so reducing energy use and decreasing
24
25 the environmental impact of wastewater discharge can also help to improve the efficiency of
26
27 wastewater treatment facilities. Water use trends analysis, water demand forecasting, and water
28
29 allocation optimization all show how AI could help to create more sustainable water
30
31 management policies (Hasan 2021). Apart from water management, artificial intelligence can
32
be used in solving many other environmental problems. Using artificial intelligence in air
33
34 pollution monitoring helps to analyze data from sensor networks and other origins to identify
35
36 origins of pollution, forecast pollution levels, and evaluate air pollution health consequences
37
38 (Agrawal et al. 2021). Through analysis of climate data, forecasting of future climate scenarios,
39
40 and evaluation of climate change impacts on different ecosystems, AI could also very much
41
42 help with climate change modeling. Species distribution data can also be analyzed by AI to
43 find threatened species and create conservation plans in environmental conservation (OECD
44
45 2022). When artificial intelligence interacts with other technologies such the Internet of Things
46
47 and sensor networks, environmental monitoring and management capabilities can be further
48
49 improved (Ullo and Sinha 2020; McClure et al. 2020). Enabled by the Internet of Things, sensor
50
51 networks would be able to gather live data on several environmental variables including soil
52
53 conditions, air quality, and water quality. Artificial intelligence algorithms would be used to
54 analyze this real-time data stream and identify and forecast pollution events, dangerous
55
56 environmental conditions, followed by timely response to minimize their impact. AI use
57
58 includes creation of early warning systems to respond to environmental threats like flooding,
59
60 droughts, and forest fires so as to allow for essential actions such as evacuation (Rozhok et al.
61
62 10
63
64
65
 2021; McGovern et al. 2022). Combining artificial intelligence with other technologies offers
1
2 us the possibility of creating very clever and adaptable environmental management systems
3
4 that could solve our planetary environmental problems. Table 2 details the machine learning
5
approaches in adsorption of contaminants.
6
7
8 Table 2: Machine Learning Approaches for Adsorption of Contaminants Using Different
9
10 Adsorbents (Abdi and Mazloom 2022; Ullah et al. 2023).
11
12 Adsorbent Contaminated ML Technique Input Variables
13
14 Target
15
16 Activated Heavy Metals (Pb, Support Vector pH, Temperature, Adsorbent
17
18 Carbon Cd) Machine Dose
19
20 Zeolite Organic Pollutants Random Forest Contact Time, Adsorbent Size,
21
22 pH
23
24
Biochar Pesticides Neural Networks Temperature, Humidity,
25 Adsorbent Dosage
26
27 Silica Gel Pharmaceuticals Decision Trees Initial Concentration, pH, Flow
28
29 Rate
30
31 Clay Dyes k-Nearest Particle Size, Adsorbent
32
33 Neighbors (k- Concentration
34
35
NN)
36 Graphene Industrial Solvents Artificial Neural Ionic Strength, Adsorbent Type
37
38 Oxide Networks
39
40 Alumina Aromatic Linear Adsorbent Dose, pH,
41
42 Compounds Regression Temperature
43
44 Activated Water Support Vector Contact Time, Ionic Strength,
45
46 Alumina Contaminants Regression Temperature
47
48
Bentonite Heavy Metals (As, Random Forest Adsorbent Concentration, pH,
49 Hg) Initial Contaminant
50
51 Concentration
52
53 Manganese Inorganic Deep Learning Adsorption Capacity, Ionic
54
55 Oxide Contaminants Networks Strength, pH
56
57
58
59
60
61
62 11
63
64
65
 Titanium Organic Pollutants Gradient Adsorbent Type, pH, Initial
1
2 Dioxide Boosting Contaminant Level
3
4 Machines
5
6
Carbon Pharmaceuticals XGBoost Temperature, Adsorbent Type,
7 Nanotubes pH
8
9 Hydrotalcite Nitrogen Support Vector Adsorbent Size, pH,
10
11 Compounds Machine Contaminant Concentration
12
13 Fly Ash Radioactive Decision Trees Particle Size, Temperature,
14
15 Materials Adsorbent Type
16
17 Modified Clay Pesticides Random Forest Adsorbent Dosage, pH, Time
18
19
20
21 4. AI-Optimized Mitigation Strategies for Endocrine Disrupting Chemicals
22
23
24 In normal circumstances, endocrine disruptions are known to compromise human and
25 environmental wellness by disrupting the hormonal system and causing several deleterious
26
27 outcomes. Some traditional EDC removal techniques are costly and ineffective; others even
28
29 generate dangerous side effects. Artificial intelligence offers great chances for perfecting
30
31 ongoing mitigation policies and creating more efficient ones (Kahn et al. 2020).
32
33
4.1 Advanced Oxidation Processes
34
35
36 With most reactive species mostly hydroxyl radicals, advanced oxidation processes
37
38 (AOPs) represent a strong set of water treatment methods that eliminate endocrine-disrupting
39
40 substances. Subtle chemical and photochemical reactions including ozonation, UV irradiation,
41
and Fenton reactions create these radicals. Non-selective hydroxyl radicals can therefore
42
43 oxidize many EDCs and change them into less toxic by-products (Vieira et al. 2021). With
44
45 artificial intelligence, AOPs could be considerably more effective, efficient, and sustainable.
46
47 Considering actual EDC levels, relevant water-quality factors (pH, temperature, etc), and
48
49 reagent properties, AI algorithms might be taught to forecast appropriate reagent dosages
50
51 required for efficient EDC removal. Avoiding the creation of toxic by-products from an
52 overdose of reagents, this feedback control system saves chemical usage, lowers treatment
53
54 expenses, and minimizes their need (Kovacic et al. 2019). Real-time optimization can also help
55
56 the AOP's performance to remain constant under different water quality and EDC
57
58 concentrations. By predicting the most beneficial reaction timing for maximum EDC removal,
59
60 therefore improving the general effectiveness of the treatment process while concurrently
61
62 12
63
64
65
 lowering the energy and time used in the treatment, AI can model the complex reaction kinetics
1
2 of AOP taking into account the EDCs' concentration, reagent quantity, water quality
3
4 parameters, and reactor design (Ao et al. 2021). The system, therefore, is dynamic, allowing
5
for the regulation of reaction time subject to actual conditions for best performance. Many
6
7 sophisticated oxidation technologies require energy input for either UV exposure or for
8
9 electrical current passing to create reactive species. By studying current ground data on EDC
10
11 levels together with reaction kinetics and water quality parameters, AI will boost energy input.
12
13 (Miklos et al. 2018). For instance, AI-assisted changes of parameters like UV intensity or
14
15 electrical current with real-time conditions will boost efforts to lower energy use, running costs,
16 and environmental effects. Intelligent control might therefore predict and avoid possible
17
18 problems like lamp fouling in UV-based AOPs, thus enhancing effectiveness and
19
20 environmental friendliness (Ghime and Ghosh 2020).
21
22
23 4.2 Adsorption and Membrane Technologies Enhanced by AI
24
25 For purging the growing pollutants from water, adsorption and membrane filtration are
26
27 proven methods. Adsorption uses substances of great surface area to draw the EDCs; membrane
28
29 filtration physically divides the EDCs based on size and other features. Including molecular
30
31 structure, polarity, and charge, artificial intelligence algorithms can evaluate the chemical
32
properties of target EDCs and forecast the most favourable adsorbent materials and/or
33
34 membrane types for a given pollutant (Gaudio et al. 2021). Used this focused strategy, one can
35
36 obtain best removal efficiency by avoiding superfluous testing, so saving material cost and
37
38 therefore lowering secondary possibility. For a more sustainable answer, artificial intelligence
39
40 can also consider the cost, availability, and environmental impact of varied materials. For
41
42 instance, it could forecast the best membrane material composition and pore size for removing
43 a certain EDC or identify fresh adsorbent materials with a very high capacity and selectivity
44
45 (Alam et al. 2022).The adsorbent materials will bind EDCs to a certain capacity, and in order
46
47 to be effective, they require periodic regeneration. Considering current data on adsorbent
48
49 performance and EDC concentrations, artificial intelligence can predict optimum regeneration
50
51 time and conditions—including temperature, pH, and regenerant concentration—thus
52
53 enhancing regeneration. This optimization lengthens the total lifetime of the adsorbents and
54 hence lowers waste generation from spent regenerants and operational expenditures (Lowe et
55
56 al. 2022). AI can predict the remaining life of the adsorbents so that they can be replaced in
57
58 advance, thus avoiding degradation in performance. Fouling in membrane filtration systems
59
60 reduces their efficiency and hence requires periodic cleaning/replacement. By analyzing sensor
61
62 13
63
64
65
 data from membrane systems that can include pressure drop, flow rate, and permeate quality,
1
2 AI can predict membrane fouling and other potential problems (Kamali et al. 2021). This
3
4 capability provides the foresight for maintaining the membrane system and limits downtime,
5
increases their lifespan, and reduces the operating costs. Further, AI can optimize the cleaning
6
7 cycle and recommend cleaning agents, improving efficiency and sustainability (Viet et al.
8
9 2022).
10
11
12 4.3 Case Studies: AI-Assisted Removal of Bisphenol A and Phthalates
13
14 Water treatment AI case studies present many of the implementation advantages and challenges
15
16 arising in real circumstances. Though large-scale pilots of AI are only now beginning to
17
18 emerge, these cases illustrate a frontier where AI can drastically enhance EDC removal
19
20 operations.
21
22 (i) Data-driven optimization of AOPs for BPA removal: Through an AI-driven optimization
23
24 of AOP process for BPA-removal, choose-H hydroxyl half-rate as reagent dosages applied
25
26 (e.g., hydrogen peroxide) or O3 and reaction-time when exposed non-phenolic like BPA
27
28 removal-more operative information win from AOPs). These phenomena were able to destroy
29
30
the BPA molecule by means of hydroxyl radicals which themselves are going to act as strong
31 oxidants in BPA removal; so, it must be really fine-tuning parameters for AOPs with respect to
32
33 a complex interaction of crucial components that favors an excess of high-level interactions
34
35 (Serna-Carrizales et al. 2024). AI can serve to adapt to this challenge by undertaking the real-
36
37 time monitoring of parameters including concentrations of BPA, water quality, and reagents
38
39 for adjusting the operational parameters of AOP. With this knowledge-based option toward the
40
41
AOP, dramatic improvement of the system for BPA removal in terms of efficiency will present
42 itself, and with that, the minimization of reagents' use and energy consumption. An example is
43
44 given where of-the-bench AI may manipulate the intensity of UV light in a UV spectrometry
45
46 AOP system based on real-time and online measurement of BPA concentration, maximizing
47
48 radical generation while shrinking unnecessary energy consumption (Abu Hasan et al. 2023).
49
50 (ii) AI-enhanced adsorption for phthalate removal: This case study demonstrates how AI
51
52 can be applied to select optimal adsorbent materials and optimize the regeneration process for
53
54 efficient removal of phthalates. Phthalates are used to confer flexibility to plastics and are
55
56 common contaminants of water (Devi et al. 2023). Adsorption shows promise for phthalate
57
58 removal, but the choice of the correct adsorbent material is paramount to successful removal.
59
AI can thus examine the chemical properties of different phthalates and predict possible
60
61
62 14
63
64
65
 adsorbent chemicals that may include activated carbon, zeolites, or biochar (Zhu et al. 2022).
1
2 In addition, AI could optimize adsorbent regeneration by predicting the optimal regeneration
3
4 time and conditions, thus maximizing adsorbent life and minimizing waste generation. As an
5
example, AI could predict the optimal temperature and pH for regenerating activated carbon
6
7 adsorbents used for phthalate removal to minimize harsh chemical use and extend adsorbent
8
9 life (Ghosh and Sahu 2022). Artificial intelligence has so much promise in the area of EDC
10
11 contamination mitigation, but there are hurdles to its application in the real world. These
12
13 include available data and data quality since AI algorithms need very large, high-quality
14
15 datasets to train and validate them. Collecting and curating such data can pose a challenge
16 given that real-world water treatment plants mostly deal with varying water sources and
17
18 fluctuating concentrations of the EDCs (Popescu et al. 2024). These implementations can also
19
20 be very resource-demanding in terms of computation, which could act as a hindrance for some
21
22 facilities. Another roadblock will be the integration of AI-based control into legacy systems,
23
24 requiring expertise. For clear and responsible application of AI in water treatment, clear
25
26 regulations will need to exist (Castelletti et al. 2023). Future research has to focus on the
27 development of even stronger AI algorithms and adaptable strategies for mitigation, as well as
28
29 input to the real-world challenges of implementation. (Chen et al. 2020), focusing on the in-
30
31 silico modelling toward the prediction of endocrine-disrupting potential, acknowledges the
32
33 growing trend of computational methods in appreciating and remedying EDC contamination.
34
35 There need to be more concerted efforts toward R&D in this regard if effective and sustainable
36
37
solutions are to be designed for mitigating EDC-related hazards. Figure 2 illustrates the flow
38 and pathways of EDCs in the environment, from production to human exposure through water
39
40 sources.
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62 15
63
64
65
 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24 Figure 2: Common pathways and origins of EDCs in the environment
25
26
27
28
29 5. Predictive Mapping of Endocrine Disrupting Chemical Contamination
30
31 Predictive mapping of endocrine-disrupting chemical contamination is essential for risk
32
33 assessment and targeted mitigation strategy development. We intend to combine geospatial
34
35 analyses with IoT sensor data and AI models to identify contamination hotspots, predict future
36
37 contamination patterns, and implement proactive measures for the protection of human and
38
39 environmental health.
40
41 5.1 Geospatial Analysis Using AI for Identifying Hotspots
42
43
44 AI-supported geospatial analysis provides a strong method for contaminated hotspot
45
46 identification through the integration of many datasets. Analysis of these spatial patterns and
47
correlations concerning the environment, land-use, demography, and hydrogeological data
48
49 leads to identifying risk areas and potential sources of contamination (Xue et al. 2022).
50
51 Environmental monitoring data from the water quality monitoring stations, which may include
52
53 soil sampling and air quality sensors, bear evidence of EDC concentrations over multiple sites,
54
55 thus helping form the basis of the understanding of the present state of contamination (Butt et
56
57 al. 2022). Land-use data can shed light on possible contamination sources. In analyzing land-
58 use patterns, AI can target areas where EDC release has a greater probability, i.e., where
59
60 industrial zones, agricultural areas, and residential areas exist. For instance, industrial discharge
61
62 16
63
64
65
 points or agricultural runoff are environmental pathways resulting in EDC contamination in
1
2 water bodies (Utami et al. 2024). Demographic data comprising population density and
3
4 socioeconomic factors and distance from possible contamination sources can assist with a finer
5
degree of risk characterization. Contaminant levels will help artificial intelligence to identify
6
7 groups under increased exposure risk, therefore focusing interventions and public health
8
9 initiatives on overlapping demographic data (Bernardini Papalia and Scognamiglio 2023). In
10
11 forecasting the distribution of pollution, hydrological data—indispensable information about
12
13 the direction of water flow, surface water bodies, and ground water resources—play a vital
14
15 part. Studies revealed water flow throughout the environment helps us to forecast the direction
16 of the EDC plumes as well as to evaluate possible effect on downstream communities and
17
18 habitats (Ehtasham et al. 2024). Given from these many datasets, artificial intelligence
19
20 algorithms are well suited for analyzing them, drawing relationships and patterns still doubted
21
22 and buried to conventional forms of analysis. With a more precise definition, it can relate EDC
23
24 levels to spatially changing variables in finding the exact hotspots and potential contamination
25
26 sources. Knowing where downstream from feasible industrial discharges or farm runoff there
27 are high EDC water levels would let for selective research and some remediation efforts (Lan
28
29 et al. 2020). (Kamel Boulos et al. 2019) draw attention to GeoAI and its importance in health
30
31 and healthcare demonstrating how location-based data and artificial intelligence might be used
32
33 to identify and grow health dangers. Studies have shown that combining GIS with AI should
34
35 provide societal advantages including pollution management and environmental monitoring.
36
37
Furthermore, using IoT sensors to transmit real-time data will more enhance the accuracy and
38 speed of the analysis, therefore permitting for proactive answer to pollution incidents as they
39
40 unfold (Ullo and Sinha 2020).
41
42
43 5.2 Integration of IoT Sensors and AI Models
44
45 The IoT sensor and AI-model-based approach for predictive mapping of EDC pollution
46
47 is indeed a noteworthy development. The IoT sensors help get high-resolution on-the-ground
48
49 data in real time with respect to EDC concentrations and concerning various environmental
50
51 matrices like water, soil, and air. Used by artificial intelligence models to more accurately
52
53 forecast and reduce, this live feed offers a dynamic and immediate evaluation of pollutants
54 levels (Popescu et al. 2024). Further boosting the accuracy of pollution maps is their constant
55
56 update and continuous improvement from real-time data gathered by IoT sensors. Generally
57
58 used, traditional monitoring techniques depend on occasional sampling and may ignore small
59
60 changes and local pollution events. Here is where IoT sensors save the day: their constant data
61
62 17
63
64
65
 flow allows for the continuous updating and better accuracy of dynamic maps showing the
1
2 contamination level at any particular time (Zheng and Paiva 2021; Buelvas et al. 2023). By the
3
4 Key enabler of accurate prediction of the spatial and temporal development of contamination
5
plumes and potential hotspot emergence, these dynamic maps help to provide perspective on
6
7 the whole perspective of likely changing contamination patterns. Real-time information from
8
9 IoT sensors is used by AI models to forecast the spread of EDC pollution (Qian and Claudel
10
11 2020). By considering weather patterns, water flow dynamics, and land use changes, these
12
13 models can predict the likely route of pollution plumes and their potential impact on
14
15 neighbouring habitats and society. The sheer power of prediction allows for timely actions:
16 changing water treatment methods or administering targeted treatments (Banjara et al. 2024).
17
18 Early warning systems rely on the synergy of real-time monitoring and predictive modelling to
19
20 be efficient. Constantly tracking EDC levels and being able to forecast next patterns of
21
22 pollution will allow an AI system to alert the pertinent actors in time—that is, citizens, water
23
24 treatment facilities, and regulatory bodies. These early notices enable rapid responses that may
25
26 stop or lessen possible damage, therefore safeguarding the environment and public health
27 (Kumar et al. 2020). For instance, a network of IoT sensors might be evenly spread across the
28
29 river basin to constantly and in real time track EDC levels at different sites. Next, the compiled
30
31 data would feed an artificial intelligence model capable of forecasting the downstream travel
32
33 of a contamination plume and delivering downstream alerts to towns and water treatment
34
35 facilities in time. These warnings will allow further proactive measures to be instituted to
36
37
minimize exposure to contaminated water, including modification of intake points and
38 temporary shut-downs (Agbehadji et al. 2023). (Popescu et al. 2024) investigates the
39
40 applications of AI and IoT in environmental pollution monitoring and management,
41
42 showcasing the capabilities of these technologies to enable the collection and analysis of real-
43
44 time data. (Alahi et al. 2023) equally discusses AI-enabled public health surveillance and the
45
46 promise of real-time data and predictive modelling for early warning systems. That is, IoT
47
48
sensor and AI model integration stands to offer powerful public health advancement as well as
49 proactiveness and effectiveness in EDC contamination management to the benefit of human
50
51 beings and the environment. Table 3 explains AI-optimized mitigation strategies for endocrine
52
53 disrupting chemicals in environmental systems.
54
55
56
Table 3: AI-Optimized Mitigation Strategies for Endocrine Disrupting Chemicals in
57 Environmental Systems (Lowe et al. 2022; Abdullah et al. 2023; Ganthavee and Trzcinski
58
59 2024).
60
61
62 18
63
64
65
 Mitigation AI Technique Endocrine Optimization Outcome
1
2 Strategy Disrupting Criteria
3
4 Chemicals
5
6
Advanced Genetic Bisphenol A, Removal Increased
7 Filtration Algorithms Phthalates Efficiency, removal
8
9 Cost efficiency
10
11 Bioremediation Neural Nonylphenol, Microbial Optimized
12
13 Enhancement Networks Atrazine Growth Rate, bioreactor
14
15 Cost design
16
17 Chemical Support Lead, Mercury Chemical Enhanced
18 Precipitation Vector Dosage, precipitation
19
20 Machine Precipitate rate
21
22 Quality
23
24 Adsorption with Random PCB, Dioxins Adsorbent Maximized
25
26 Carbon-Based Forest Type, Contact adsorption
27
28 Materials Time capacity
29
30
Electrochemical Deep Learning Pesticides, Power Improved
31 Treatment Perfluorinated Consumption, energy
32
33 Compounds Efficiency efficiency
34
35 Photocatalytic Reinforcement BPA, Triclosan Light Optimized
36
37 Degradation Learning Intensity, degradation
38
39 Catalytic rate
40
41
Activity
42 Membrane K-Nearest Estrogen, PCBs Flow Rate, Enhanced
43
44 Filtration Neighbors (K- Membrane filtration
45
46 NN) Pore Size performance
47
48 Phytoremediation Decision Trees Endosulfan, DDT Plant Growth, Improved
49
50 Chemical contaminant
51
52
Uptake uptake
53 Ultrafiltration Artificial Chlorinated Pressure, Optimized
54
55 Neural Compounds Membrane filtration
56
57 Networks Material efficiency
58
59
60
61
62 19
63
64
65
 Electrocoagulation Gradient Organophosphates, Coagulant Enhanced
1
2 Boosting Pesticides Type, Voltage coagulation
3
4 Machines process
5
6
Biological Random Alkylphenols, Filter Media, Enhanced
7 Filtration Forest Pesticides Bacterial microbial
8
9 Activity degradation
10
11 Magnetic Support Heavy Metals (Cd, Magnetic Improved
12
13 Separation Vector Pb) Field, Contact metal
14
15 Machines Time removal
16
17 Supercritical Fluid Deep Learning Phthalates, Fluid Pressure, Optimized
18 Extraction Benzophenones Temperature extraction
19
20 efficiency
21
22 Adsorption with Neural Hormones, Adsorbent Increased
23
24 Metal Oxides Networks Herbicides Concentration, adsorption
25
26 pH rate
27
28 Enzymatic Genetic Triclosan, DDT Enzyme Type, Enhanced
29
30
Treatment Algorithms Reaction Time degradation
31 rate
32
33 Water Treatment Reinforcement Endocrine Clay Dosage, Optimized
34
35 with Clays Learning Disrupting pH, Contact pesticide
36
37 Pesticides Time removal
38
39
40
41 6. Quantitative and Qualitative Outcomes
42
43
44 6.1 Model Accuracy, Sensitivity, and Specificity
45
46 Key metrics have to be evaluated to assess the AI performances in reducing the EDCs. The key
47
48 metrics provide a quantitative measure of how well the predicting and classification functions
49
50 of a given model correspond to EDC contamination events. The accuracy metric is a balance
51
52 of the positive and negative predictions done by the model. It gives the percentage of instances
53
54 that have been correctly classified, combining true positives, the identification of
55
contamination events that are actually present, and true negatives, the non-identification of
56
57 contamination events that are actually not present. High accuracy is alarmingly crucial if EDC
58
59 detection and prediction are to be reliable, minimizing the chances of false positives and false
60
61
62 20
63
64
65
 negatives. However, accuracy can be deceptive when one class in an impaired dataset
1
2 (contamination events) is normally kept well out of a majority class (Hachmi et al. 2019;
3
4 Collins and Barton-Maclaren 2022). Sensitivity (also known as recall or true positive rate)
5
assesses the ability of a model to correctly identify true positive cases. It is the proportion of
6
7 actual EDC contamination events that the model identifies correctly. High sensitivity is
8
9 important for minimizing false negatives, which are the missed contamination events. In EDCs,
10
11 false negatives are prioritized for minimization due to the health and environmental impact of
12
13 missed contamination. A model with very high sensitivity generally ensures that all potential
14
15 contamination events are picked up, even at the expense of higher rates of false positives
16 (Monaghan et al. 2021). Specificity evaluates a model's ability to correctly identify true
17
18 negative cases. More specifically, it measures the class of non-contamination events correctly
19
20 classified by the model. High specificity is important for minimizing false positives, which are
21
22 cases in which the model predicts contamination, but there really is none. Contrary to false
23
24 negatives, false positives do run the risk of leading to unnecessary interventions, spending of
25
26 resources, and potential disruptions (Al-Battbootti et al. 2024).
27
28 The particular target values for accuracy, sensitivity, and specificity may depend on
29
30 application-related considerations and the relative costing of false-positives and false-
31
32 negatives. For example, where missing a contamination event imposes heavy penalties, high
33
34 sensitivity might be adopted at the cost of specificity (Grunau and Linn 2018). Conversely, if
35 particular false positives are severely penalized, one must try for great specificity. Balancing
36
37 these measures entails thoughtful consideration of the particular uses as well as the
38
39 compromise. Other factors to take into account are metrics including the F1 score, which
40
41 considers precision and recall. Other techniques that can help to evaluate the model's
42
43 performance on various thresholds so that an ideal compromise between sensitivity and
44
45
specificity is selected are the use of Receiver Operating Characteristic curves and AUC
46 (Yacouby and Axman 2020).
47
48
49 6.2 Reduction Efficiency of AI-Enhanced Mitigation Technologies
50
51
EDCs are evaluated for actual attenuation using AI by measuring pollution reduction as
52
53 a decrease in EDC concentration (Bhagat et al. 2020). The efficiency of removal varies
54
55 depending on the technology. AOPs use powerful oxidants to destroy EDCs; membrane
56
57 filtration physically separates them by size and charge; and bioremediation utilizes
58
59 microorganisms for metabolism (Vieira et al. 2021). The type of EDC, target reduction levels,
60
61
62 21
63
64
65
 cost, and infrastructure all have a role in selecting an acceptable technique. Hybrid techniques
1
2 can improve removal efficiency (Tufail et al., 2021). AI-driven mitigation solutions are
3
4 optimized by adjusting treatment time, flow rate, oxidant concentration in AOPs, and
5
membrane characteristics (Viet et al. 2022). AI continuously optimizes these parameters based
6
7 on sensor and process monitoring inputs to improve efficacy while lowering energy and
8
9 operational costs (Suthar et al. 2021). Environmental parameters like as temperature, pH, and
10
11 co-occurring pollutants all have an impact on performance. Temperature affects the rates of
12
13 AOP reactions as well as microbial activity in bioremediation, while pH regulates EDC
14
15 speciation and membrane performance (Giwa et al. 2021). Other co-occurring pollutants can
16 disrupt the occupancy of reaction sites, affecting elimination. AI predicts and optimizes
17
18 treatment based on these variables to ensure robustness under changing situations. In situ
19
20 monitoring is conceivable, allowing adaptive AI control to improve efficiency (Popescu et al.,
21
22 2024). AI-driven EDC reduction is assessed for cost-effectiveness in investment and policy
23
24 decisions. Life Cycle Assessment and Cost-Benefit Analysis improve the sustainability and
25
26 economic viability of these activities (Pesqueira et al., 2020). Figure 3 illustrates different AI
27 models and their applications in environmental and industrial processes.
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55 Figure 3: Applications of AI Models in Environmental and Industrial Systems
56
57
58
59
60 7. Challenges and Future Directions
61
62 22
63
64
65
 AI has made substantial contributions to EDC monitoring and reduction, but certain
1
2 constraints must be overcome before it can fully realize its potential. The most significant
3
4 difficulty is the reliance on huge, high-quality datasets that are frequently inconsistent, partial,
5
or scarce, compromising model robustness (Comess et al. 2020). Data noise, outliers, and
6
7 missing values further deteriorate model performance, necessitating the implementation of
8
9 consistent monitoring programs and improved data-cleaning procedures (Whang et al. 2023).
10
11 Furthermore, AI models, particularly those based on deep learning, are "black box" in nature,
12
13 which limits interpretability and regulatory acceptance. Creating explainable AI techniques can
14
15 improve transparency and trust in AI-powered EDC monitoring (Paudyal and Wong 2018).
16 Another issue is weak generalizability, which means that models trained on a single dataset
17
18 may fail in multiple environmental conditions. Transfer learning and domain adaption
19
20 approaches can boost adaptability (Beltiukov et al. 2023). Furthermore, AI training and
21
22 implementation need significant processing capacity, which limits accessibility. Efficient
23
24 algorithms, cloud computing, and edge computing technologies can help to address this issue
25
26 (Aldoseri et al. 2023). Despite these obstacles, AI has numerous potential, including improved
27 models such as deep reinforcement learning and hybrid approaches that improve accuracy and
28
29 flexibility (Naik et al. 2022). Multi-omics integration, which combines genomes,
30
31 transcriptomics, proteomics, and metabolomics, enables a comprehensive understanding of
32
33 EDC impacts and early warning systems (Haider et al. 2020). IoT-connected sensors provide
34
35 for real-time monitoring and automatic risk reduction (Ullo & Sinha 2020). Crowdsourced data
36
37
and citizen science can help to expand EDC surveillance, while AI improves data processing
38 and exposure mapping (McClure et al. 2020). Addressing present constraints while utilizing
39
40 AI's promise would transform EDC monitoring, resulting in improved environmental and
41
42 human health protection.
43
44
45 8. Conclusion
46
47 This study on artificial intelligence predictive models reveals a transformation ability
48
49 of artificial intelligence for solving this serious environmental issue by means of mapping and
50
51 endocrine disruptors in water systems. By effectively predicting the presence and distribution
52
53 and fate of EDCs in water ecosystems, artificial intelligence driven predictive modeling helps
54 to find pollution hotspots, conduct risk evaluation of human and environmental health, and plan
55
56 specific damage control policies. Along with online AI app, EDC monitoring and management
57
58 point the most important way for solving this serious environmental problem. Using machine
59
60 learning, big data analytics, and cutting-edge sensors, we can develop more efficient methods
61
62 23
63
64
65
 of spotting, forecasting, and eliminating endocrine disrupting chemicals from all water
1
2 systems. These developments could go further to enhance existing water treatment
3
4 technologies or lead to viable options for new development in remediation. AI models can
5
integrate real-time sensor data into IoT platforms, and this will give rise to smart water
6
7 management systems that will allow for dynamic adaptation to changing conditions concerning
8
9 the delivery of safe and sustainable water resources. With challenges in availability of data,
10
11 model interpretability, and computational costs, this continues to be an evolving area of
12
13 research and much is being done to develop AI that will conceptually protect water resources
14
15 from the harmful effects of EDCs. To realize this potential and create a healthier world that is
16 more sustainable, collaborative efforts across scientific disciplines and investments in AI-
17
18 supported solutions must continue.
19
20
21 Funding Declaration:
22
23 This study was conducted without any financial support.
24
25
26
27
28 Competing Interests Declaration:
29
30
31 The authors declare no competing interests.
32
33
34
Author Contributions:
35
36  V S: Conceptualization, supervision, and final manuscript review.
37
38
39
 S S I: Literature review, writing-original draft.
40
41  S. A: Data collection, analysis, and manuscript editing.
42
43
44  B. S. B: Writing-review & editing, formatting.
45
46  G G: Methodology, validation, and critical revisions.
47
48
49  H C Supervision, Writing review and editing.
50
51  TM : Supervision, Writing review and editing
52
53
54 References
55
56 Abaineh A, Ejigu D, Atlabachew M, et al (2024) Risks of pesticides on aquatic ecosystems and
57
58 pesticide management effectiveness in Ethiopia: review. Int J Environ Sci Technol 21:8833–
59 8848. [Link]
60
61
62 24
63
64
65
 Abdi J, Mazloom G (2022) Machine learning approaches for predicting arsenic adsorption from
1 water using porous metal–organic frameworks. Sci Rep 12:16458.
2
3 [Link]
4
5 Abdullah MS, Goh PS, Ismail AF, Hasbullah H (2023) The Treatment of Endocrine-Disruptive
6 Chemicals in Wastewater through Asymmetric Reverse Osmosis Membranes: A Review.
7 Symmetry 15:1049. [Link]
8
9 Abu Hasan H, Muhamad MH, Budi Kurniawan S, et al (2023) Managing Bisphenol A
10
11 Contamination: Advances in Removal Technologies and Future Prospects. Water 15:3573.
12 [Link]
13
14 Agbehadji IE, Mabhaudhi T, Botai J, Masinde M (2023) A Systematic Review of Existing Early
15 Warning Systems’ Challenges and Opportunities in Cloud Computing Early Warning Systems.
16
17 Climate 11:188. [Link]
18
19 Agrawal P, Sinha A, Kumar S, et al (2021) Exploring Artificial Intelligence Techniques for
20 Groundwater Quality Assessment. Water 13:1172. [Link]
21
22 Ahmed U, Mumtaz R, Anwar H, et al (2020) Water quality monitoring: from conventional to
23 emerging technologies. Water Supply 20:28–45. [Link]
24
25 Ahn C, Jeung E-B (2023) Endocrine-Disrupting Chemicals and Disease Endpoints. IJMS
26
27 24:5342. [Link]
28
29 Alahi MEE, Sukkuea A, Tina FW, et al (2023) Integration of IoT-Enabled Technologies and
30 Artificial Intelligence (AI) for Smart City Scenario: Recent Advancements and Future Trends.
31 Sensors 23:5206. [Link]
32
33 Alam G, Ihsanullah I, Naushad Mu, Sillanpää M (2022) Applications of artificial intelligence
34
35
in water treatment for optimization and automation of adsorption processes: Recent advances
36 and prospects. Chemical Engineering Journal 427:130011.
37 [Link]
38
39 Al-Battbootti MJH, Marin I, Al-Hameed S, et al (2024) Designing and Developing an
40
41
Advanced Drone-Based Pollution Surveillance System for River Waterways, Streams, and
42 Canals Using Machine Learning Algorithms: Case Study in Shatt al-Arab, South East Iraq.
43 Applied Sciences 14:2382. [Link]
44
45 Almazrouei B, Islayem D, Alskafi F, et al (2023) Steroid hormones in wastewater: Sources,
46
47
treatments, environmental risks, and regulations. Emerging Contaminants 9:100210.
48 [Link]
49
50 Ao X, Eloranta J, Huang C-H, et al (2021) Peracetic acid-based advanced oxidation processes
51 for decontamination and disinfection of water: A review. Water Research 188:116479.
52
[Link]
53
54
55
Auyb MA, Zamir MT, Khan I, et al (2024) Social Media and Artificial Intelligence for
56 Sustainable Cities and Societies: A Water Quality Analysis Use-case
57
58
59
60
61
62 25
63
64
65
 Azizi D, Arif A, Blair D, et al (2022) A comprehensive review on current technologies for
1 removal of endocrine disrupting chemicals from wastewaters. Environmental Research
2
3 207:112196. [Link]
4
5 Banjara M, Bhusal A, Ghimire AB, Kalra A (2024) Impact of Land Use and Land Cover
6 Change on Hydrological Processes in Urban Watersheds: Analysis and Forecasting for Flood
7 Risk Management. Geosciences 14:40. [Link]
8
9 Bernardini Papalia R, Scognamiglio G (2023) Environmental justice: geostatistical analysis of
10
11 environmental hazards and socioeconomic factors—the case of Italy. GeoJournal 88:6221–
12 6246. [Link]
13
14 Buelvas J, Múnera D, Tobón V. DP, et al (2023) Data Quality in IoT-Based Air Quality
15 Monitoring Systems: a Systematic Mapping Study. Water Air Soil Pollut 234:248.
16
17 [Link]
18
19 Butt MA, Voronkov GS, Grakhova EP, et al (2022) Environmental Monitoring: A
20 Comprehensive Review on Optical Waveguide and Fiber-Based Sensors. Biosensors 12:1038.
21 [Link]
22
23 Castelletti A, Ficchì A, Cominola A, et al (2023) Model Predictive Control of water resources
24
25
systems: A review and research agenda. Annual Reviews in Control 55:442–465.
26 [Link]
27
28 Chen Q, Zhou C, Shi W, et al (2020) Mechanistic in silico modeling of bisphenols to predict
29 estrogen and glucocorticoid disrupting potentials. Science of The Total Environment
30
31
728:138854. [Link]
32
33
Collins SP, Barton-Maclaren TS (2022) Novel machine learning models to predict endocrine
34 disruption activity for high-throughput chemical screening. Front Toxicol 4:981928.
35 [Link]
36
37 Devi T, Saleh NM, Kamarudin NHN, et al (2023) Efficient adsorption of organic pollutants
38
39
phthalates and bisphenol A (BPA) utilizing magnetite functionalized covalent organic
40 frameworks (MCOFs): A promising future material for industrial applications. Ecotoxicology
41 and Environmental Safety 268:115706. [Link]
42
43 Ehtasham L, Sherani SH, Nawaz F (2024) Acceleration of the hydrological cycle and its impact
44
45
on water availability over land: an adverse effect of climate change. Meteorol Hydrol Water
46 Manage. [Link]
47
48 Gałązka A, Jankiewicz U (2022) Endocrine Disrupting Compounds (Nonylphenol and
49 Bisphenol A)–Sources, Harmfulness and Laccase-Assisted Degradation in the Aquatic
50 Environment. Microorganisms 10:2236. [Link]
51
52
Ganthavee V, Trzcinski AP (2024) Artificial intelligence and machine learning for the
53
54 optimization of pharmaceutical wastewater treatment systems: a review. Environ Chem Lett
55 22:2293–2318. [Link]
56
57 Gaston SA, Birnbaum LS, Jackson CL (2020) Synthetic Chemicals and Cardiometabolic
58 Health Across the Life Course Among Vulnerable Populations: a Review of the Literature from
59
60 2018 to 2019. Curr Envir Health Rpt 7:30–47. [Link]
61
62 26
63
64
65
 Gaudio MT, Coppola G, Zangari L, et al (2021) Artificial Intelligence-Based Optimization of
1 Industrial Membrane Processes. Earth Syst Environ 5:385–398.
2
3 [Link]
4
5 Ghime D, Ghosh P (2020) Advanced Oxidation Processes: A Powerful Treatment Option for
6 the Removal of Recalcitrant Organic Compounds. In: Bustillo-Lecompte C (ed) Advanced
7 Oxidation Processes - Applications, Trends, and Prospects. IntechOpen
8
9 Ghosh S, Sahu M (2022) Phthalate pollution and remediation strategies: A review. Journal of
10
11 Hazardous Materials Advances 6:100065. [Link]
12
13 Gonsioroski A, Mourikes VE, Flaws JA (2020) Endocrine Disruptors in Water and Their
14 Effects on the Reproductive System. IJMS 21:1929. [Link]
15
16 Grunau G, Linn S (2018) Commentary: Sensitivity, Specificity, and Predictive Values:
17 Foundations, Pliabilities, and Pitfalls in Research and Practice. Front Public Health 6:256.
18
19 [Link]
20
21 Grzegorzek M, Wartalska K, Kowalik R (2024) Occurrence and sources of hormones in water
22 resources—environmental and health impact. Environ Sci Pollut Res 31:37907–37922.
23 [Link]
24
25 Hachmi F, Boujenfa K, Limam M (2019) Enhancing the Accuracy of Intrusion Detection
26
27 Systems by Reducing the Rates of False Positives and False Negatives Through Multi-
28 objective Optimization. J Netw Syst Manage 27:93–120. [Link]
29 9459-y
30
31 Hasan MA (2021) An emergent addition for the optimal systemization of wastewater utilization
32
33
plants using artificial intelligence. Water Science and Technology 84:2805–2817.
34 [Link]
35
36 Hino M, Benami E, Brooks N (2018) Machine learning for environmental monitoring. Nat
37 Sustain 1:583–588. [Link]
38
39 Hmoud Al-Adhaileh M, Waselallah Alsaade F (2021) Modelling and Prediction of Water
40
41
Quality by Using Artificial Intelligence. Sustainability 13:4259.
42 [Link]
43
44 Jones RR, Stavreva DA, Weyer PJ, et al (2020) Pilot study of global endocrine disrupting
45 activity in Iowa public drinking water utilities using cell-based assays. Science of The Total
46
47
Environment 714:136317. [Link]
48
49
Kahn LG, Philippat C, Nakayama SF, et al (2020) Endocrine-disrupting chemicals:
50 implications for human health. The Lancet Diabetes & Endocrinology 8:703–718.
51 [Link]
52
53 Kamali M, Appels L, Yu X, et al (2021) Artificial intelligence as a sustainable tool in
54
55
wastewater treatment using membrane bioreactors. Chemical Engineering Journal 417:128070.
56 [Link]
57
58
59
60
61
62 27
63
64
65
 Kamel Boulos MN, Peng G, VoPham T (2019) An overview of GeoAI applications in health
1 and healthcare. Int J Health Geogr 18:7, s12942-019-0171–2. [Link]
2
3 019-0171-2
4
5 Kavianpour B, Piadeh F, Gheibi M, et al (2024) Applications of artificial intelligence for
6 chemical analysis and monitoring of pharmaceutical and personal care products in water and
7 wastewater: A review. Chemosphere 368:143692.
8
9 [Link]
10
11 Kovacic M, Kusic H, Loncaric Bozic A, Dionysiou DD (2019) Advanced Oxidation Processes.
12 In: Maurice P (ed) Encyclopedia of Water, 1st edn. Wiley, pp 1–15
13
14 Kumar M, Sarma DK, Shubham S, et al (2020) Environmental Endocrine-Disrupting Chemical
15 Exposure: Role in Non-Communicable Diseases. Front Public Health 8:553850.
16
17 [Link]
18
19 Lan Y, Tang W, Dye S, Delmelle E (2020) A web-based spatial decision support system for
20 monitoring the risk of water contamination in private wells. Annals of GIS 26:293–309.
21 [Link]
22
23 Li L, Rong S, Wang R, Yu S (2021) Recent advances in artificial intelligence and machine
24
25
learning for nonlinear relationship analysis and process control in drinking water treatment: A
26 review. Chemical Engineering Journal 405:126673. [Link]
27
28 Lowe M, Qin R, Mao X (2022) A Review on Machine Learning, Artificial Intelligence, and
29 Smart Technology in Water Treatment and Monitoring. Water 14:1384.
30
31
[Link]
32
33
McClure EC, Sievers M, Brown CJ, et al (2020) Artificial Intelligence Meets Citizen Science
34 to Supercharge Ecological Monitoring. Patterns 1:100109.
35 [Link]
36
37 McGovern A, Ebert-Uphoff I, Gagne DJ, Bostrom A (2022) Why we need to focus on
38
39
developing ethical, responsible, and trustworthy artificial intelligence approaches for
40 environmental science. Environ Data Science 1:e6. [Link]
41
42 Metcalfe CD, Bayen S, Desrosiers M, et al (2022) Methods for the analysis of endocrine
43 disrupting chemicals in selected environmental matrixes. Environmental Research
44
45
206:112616. [Link]
46
47
Miklos DB, Remy C, Jekel M, et al (2018) Evaluation of advanced oxidation processes for
48 water and wastewater treatment – A critical review. Water Research 139:118–131.
49 [Link]
50
51 Mishra RK, Mentha SS, Misra Y, Dwivedi N (2023) Emerging pollutants of severe
52
environmental concern in water and wastewater: A comprehensive review on current
53
54 developments and future research. Water-Energy Nexus 6:74–95.
55 [Link]
56
57 Mitra S, Saran RK, Srivastava S, Rensing C (2024) Pesticides in the environment: Degradation
58 routes, pesticide transformation products and ecotoxicological considerations. Science of The
59
60 Total Environment 935:173026. [Link]
61
62 28
63
64
65
 Monaghan TF, Rahman SN, Agudelo CW, et al (2021) Foundational Statistical Principles in
1 Medical Research: Sensitivity, Specificity, Positive Predictive Value, and Negative Predictive
2
3 Value. Medicina 57:503. [Link]
4
5 OECD (2022) Measuring the environmental impacts of artificial intelligence compute and
6 applications: The AI footprint
7
8 Papalou O, Kandaraki EA, Papadakis G, Diamanti-Kandarakis E (2019) Endocrine Disrupting
9 Chemicals: An Occult Mediator of Metabolic Disease. Front Endocrinol 10:112.
10
11 [Link]
12
13 Popescu SM, Mansoor S, Wani OA, et al (2024) Artificial intelligence and IoT driven
14 technologies for environmental pollution monitoring and management. Front Environ Sci
15 12:1336088. [Link]
16
17 Qian K, Claudel C (2020) Real-time mobile sensor management framework for city-scale
18
19 environmental monitoring. Journal of Computational Science 45:101205.
20 [Link]
21
22 Rehman K, Fatima F, Waheed I, Akash MSH (2018) Prevalence of exposure of heavy metals
23 and their impact on health consequences. J Cell Biochem 119:157–184.
24
25
[Link]
26
27 Rozhok AP, Storozhenko AS, Valiaeva AV, et al (2021) Methods of monitoring the Ground-
28 Climate-Pipeline system in sections with hazardous processes. IOP Conf Ser: Earth Environ
29 Sci 864:012022. [Link]
30
31 Serna-Carrizales JC, Zárate-Guzmán AI, Flores-Ramírez R, et al (2024) Application of
32
33
artificial intelligence for the optimization of advanced oxidation processes to improve the water
34 quality polluted with pharmaceutical compounds. Chemosphere 351:141216.
35 [Link]
36
37 Sharma A, Mollier J, Brocklesby RWK, et al (2020) Endocrine‐disrupting chemicals and male
38
39
reproductive health. Reprod Medicine & Biology 19:243–253.
40 [Link]
41
42 Stiefel C, Stintzing F (2023) Endocrine-active and endocrine-disrupting compounds in food –
43 occurrence, formation and relevance. NFS Journal 31:57–92.
44
45
[Link]
46
47
Tang Z, Liu Z, Wang H, et al (2020) Trace determination of eleven natural estrogens and
48 insights from their occurrence in a municipal wastewater treatment plant and river water. Water
49 Research 182:115976. [Link]
50
51 Ullah H, Khan S, Chen B, et al (2023) Machine learning approach to predict adsorption
52
capacity of Fe-modified biochar for selenium. Carbon Res 2:29.
53
54 [Link]
55
56 Ullo SL, Sinha GR (2020) Advances in Smart Environment Monitoring Systems Using IoT and
57 Sensors. Sensors 20:3113. [Link]
58
59
60
61
62 29
63
64
65
 Utami W, Sugiyanto C, Rahardjo N (2024) Artificial intelligence in land use prediction
1 modeling: a review. IJ-AI 13:2514. [Link]
2
3 Vieira WT, De Farias MB, Spaolonzi MP, et al (2021) Latest advanced oxidative processes
4
5 applied for the removal of endocrine disruptors from aqueous media – A critical report. Journal
6 of Environmental Chemical Engineering 9:105748. [Link]
7
8 Viet ND, Jang D, Yoon Y, Jang A (2022) Enhancement of membrane system performance using
9 artificial intelligence technologies for sustainable water and wastewater treatment: A critical
10
11 review. Critical Reviews in Environmental Science and Technology 52:3689–3719.
12 [Link]
13
14 Xue J, Zartarian V, Tornero-Velez R, et al (2022) A Generalizable Evaluated Approach,
15 Applying Advanced Geospatial Statistical Methods, to Identify High Lead Exposure Locations
16
17 at Census Tract Scale: Michigan Case Study. Environ Health Perspect 130:077004.
18 [Link]
19
20 Yacouby R, Axman D (2020) Probabilistic Extension of Precision, Recall, and F1 Score for
21 More Thorough Evaluation of Classification Models. In: Proceedings of the First Workshop on
22
23
Evaluation and Comparison of NLP Systems. Association for Computational Linguistics,
24 Online, pp 79–91
25
26 Zheng H, Paiva A (2021) Assessing Machine Learning Approaches to Address IoT Sensor Drift
27
28 Zhong S, Zhang K, Bagheri M, et al (2021) Machine Learning: New Ideas and Tools in
29 Environmental Science and Engineering. Environ Sci Technol [Link].1c01339.
30
31
[Link]
32
33
Zhu Z, Rao R, Zhao Z, et al (2022) Research progress on removal of phthalates pollutants from
34 environment. Journal of Molecular Liquids 355:118930.
35 [Link]
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62 30
63
64
65