Machine Learning Enabled Smart IoT Framework for

Health Monitoring
Akanksha Tiwari1, Pradeep Jha2, Pankaj Jain3, Amrit Pal Kaur4, Shivam Kumar5
1,2,3,4,5
Department of CSE, Global Institute of Technology, Jaipur, Rajasthan, India
akankshatiwari.04cs@[Link], pradeep.jha1988@[Link], [Link]@[Link],
23egjcs201@[Link], 23egjcs205@[Link]

Keywords: IoT, Healthcare, Machine Learning, SPO2, SVM, Accuracy.

Abstract
The accelerated digitalization of the healthcare sector has resulted in the pace of the evolution of smart systems
to track the patients in real-time and distance. IoT devices like a wearable sensor and connected medical
equipment generate colossal volumes of real-time physiology data that hold colossal capacities of giving early
diagnosis of the diseases. The given work proposes Smart IoT-Based Health Monitoring Framework with the
assistance of which the vital signs heart rate, body temperature and oxygen saturation (SpO2) will be assessed
with references to Machine Learning (ML). The acquired data is pre-processed and tested using the assistance of
Logistic Regression, Support Vector machine (SVM) and Random Forest algorithms. The experiment made on a
publicly available physiological dataset indicates that the most effective model is the Random Forest model
which has the best performance with the accuracy of 95.68 and good values of precision and recall. The
suggested layered architecture will be an integration of sensor-level data application, cloud-based processing,
ML-based forecasting, and an automatic alert system to recognize anomalies in time. The findings signify the
plausibility, scalability, and applicability of the ML-based IoT systems in modern healthcare. The second step of
work will be to implement the techniques of deep learning and enhanced IOT communication protocols that will
add to the further increase in predictive performance and flexibility.

1 Introduction systems however is highly indicative of the capability

of such information to be processed right and to filter
Advances in technology that allow performing a noise, identify trends and irregularities on time [9],
checkup repeatedly and analyzing data in real time [10].
and the individual modulation of health have a high
pace, transforming the world healthcare system [1].
The past health assessment process relies majorly on
the periodic clinical examination, physical
measurements and symptomatic diagnosis [2].
Although effective in a way, they do not offer an
unceasing flow of information regarding a
physiological condition of a patient and frequently
lead to a late diagnosis of a critical disease [3]. Next
to the rise of lifestyle-associated diseases, heart-
related disorders, respiratory complications,
pandemics, such as the COVID-19 epidemic, there is
a high demand of intelligent systems, which will be
able to exercise unremitting control and early
warning of deviations [4].
The last few years have witnessed the purpose of the
Internet of Things (IoT) in the healthcare industry
that allows smart sensors and devices to monitor,
transmit, and store information related to patients in
large amounts [5], [6]. The constantly streaming
streams of physiology information produced by
wearable machines pulse oximeters, heart rate FIG 1: IoT Enabled health monitoring architecture
monitors, fitness watches, even smart thermometers,
The intelligent method of managing physiological
and, consequently, permit remote monitoring, even in
data with the support of the sophisticated data
circumstances where there is no hospital
analysis capabilities is known as Machine Learning
infrastructure available [7], [8]. The value of such
(ML). ML models are able to detect small tendencies
of health parameters that an individual will not
recognize manually. The models are trained on the
past data to be conversant with the difference
between normal and abnormal health condition and
can be used in future on the real-time prediction [11],
[12].
This paper will present a Smart IoT-Based Health
Monitoring Framework, which is driven by Machine
Learning, bringing together wearable sensor data and
subsequent automated classification algorithms. The
system tracks three vital health parameters heart rate
(HR), oxygen saturation (SpO 2), and body
temperature because changes in each of them may
indicate infections, cardiovascular problems,
respiratory health, and metabolism disorders. The
framework then classifies the health condition of a
user as either normal or abnormal using the Logistic
Regression, Support Vector Machine (SVM), and
Randome Forest classifiers. It is primarily aimed at
creating a low-cost, scalable, and user-friendly Fig 2: Flowchart of proposed methodology
solution that would be able to accommodate
individuals, healthcare professionals, and B. Data Preprocessing
telemedicine platforms. The system allows To obtain quality and consistency of data, several
identifying anomalies in real-time, minimizing the pre-processing processes are undertaken:
workload in the hospital, and facilitating constant
observation of those of advanced age, chronic  Noise Reduction: A moving average smoothing
patients, and groups of people in remote locations. filter is applied in order to remove high
The paper also deals with the main challenges of the frequency noise and stabilize sensor values.
IoT-based healthcare, such as sensor reliability, data  Missing Data: There is also an inconvenient data
privacy, communication latency, and computer loss due to network hiccups which is countered
demand. The proposed framework helps in improving with the aid of mean substitution in a bid to
the future of intelligent, secure, and next-generation provide continuity in time-series.
digital health care systems by using predictive  Feature Scaling: Physiological data is not equal
analytics to manage noisy and multidimensional data. and various measures have varying ranges
therefore to normalize all features within the
2 Proposed Technique range of [0, 1], we use Min-Max normalization
to normalize all features to the range between 0
The proposed Smart IoT-Based Health monitor and 1 to ensure better machine learning models.
Framework is designed and has a methodology that
C. Feature Extraction
comprises of data acquisition stage, preprocessing
stage, feature extraction stage, machine learning- It is the extraction of features that establishes
based classification stage, model evaluation stage and meaning patterns of the raw physiological signals:
real-time alert generation stage. It is a combined  ECG and Heart Rate measurements: The
working solution that ensures an effective processing parameters obtained are the Root Mean Square
of physiological signals and enables remote health (RMS), and Heart Rate Variability (HRV)
control. The flowchart of the proposed methodology parameters that are determined to assess cardiac
is shown in the Fig 2. health tendencies.
 Temperature and SpO2 Characteristics:
A. Data Acquisition Layer Statistic mean, variance and coefficient of
Fitbit IoT sensors continuously measure real-time variation is computed to make a comparison
physiological measurements (heart rate, blood between normal physiological conditions and
pressure, oxygen saturation, SpO2, body temperature, abnormal conditions.
ECG). These sensors have results in the form of time- These extracted features are fed to classification
series data streams delivered to the cloud in models.
lightweight communications protocols such as MQTT
or HTTP. This ensures that there is sound D. Machine Learning Classifier
communication between the cloud layers and the The supervised learning algorithm is trained in such a
device with low latency. way that it recognizes the physiological conditions to
be either normal or abnormal:
 Logistic Regression: It is a linear classifier that also revealed that the false negatives of the Random
is based on a baseline technique, and it is applied Forest model were the fewest and it is highly
to determine the likelihood of anomalies. important in health monitoring situations where
 Support Vector Machine (SVM): This forms instances of undetectable abnormal cases may lead to
an ideal decision-making line that is effective an enormous medical consequence. The correctness
especially in scenarios where there is a and memory of the abnormal category was over 94%
physiological overlap. and the reliability was quite robust in forecasting the
 Random Forest: It represents an ensemble vital health abnormalities.
model that entails the integration of a number of
decision trees together with the majority voting Table 1: Table of performance comparison
system, which is especially suitable in testing Accuracy F1-
Model Precision Recall
nonlinear and noisy biomedical data. (%) Score
The hyper parameters are optimized through the Logistic
89.24 0.88 0.87 0.87
Regression
cross-validation in order to achieve better
SVM 92.14 0.91 0.92 0.92
performance.
Random
95.68 0.96 0.95 0.95
E. Model Evaluation Forest
The performance measures that are used to measure
the classifiers are traditional performance measures The high F1-score and the fact that the model is
like accuracy, precision, recall and F1-score. The balanced in the two classes also testify to the nature
evaluation of the classification errors and sensitivity of the model being balanced. The comparison
to the abnormal physiological conditions is also between the measures of evaluation indicates that the
carried out by the confusion matrix analysis. Random Forest is statistically superior to the other
two models on all of the performance measures and,
F. Live Alerts and Implementation. thus, better adapted to the biomedical classification
The trained model is called on a cloud based tasks in real life. Moreover, the model has been
inference engine. The alerts are sent immediately to compared during cross validation with the
the mobile application of the user and hospital presentation of a low variation of facade of folds and,
dashboard in case of abnormal patterns. This enables consequently, a high generalization.
medical intervention and clinical decision making.

3 Results and Discussion

It was confirmed through the testing and training of
the three machine-learning classifiers, which included
Logistic Regression, Support Vector Machine (SVM),
and Random Forest, being used on the process
physiological data in order to test the proposed Smart
IoT-Enabled Health Monitoring Framework. The
analysis will focus on the accuracy of prediction,
prediction by classes and generally stability of the
model. The experiments were executed using the
same dataset dividing it into parts so that a fair
comparison could be made. To compare all of the
models on a rigorous level, such typical measures as
accuracy, precision, recall, F1-score, and the findings
of the confusion matrix were computed. This was the
maximum predictive power of the classifier of the
Random Forest, which reached the value of 95.68 and Fig 3: Confusion Matrix of RFC
was very high compared with the values of Logistic
Regression and SVM. The superiority of the Random Besides the performance of the model, the test on
Forest is attributed to the fact that the ensemble form applicability of the trained Random Forest classifier
eradicates variance, non-linear relations and is useful to real-time was also performed in the context of the
in mitigating the impact of noise that arises in the IoT. The rate of latency by which sensor information
physiological measurements. On the other side there was being relayed by the wearables was extremely
was a less accuracy of the Logistic Regression due to low and the cloud inference layer would practically
its linear decision making model as compared to the require no time to make a prediction. The alert
SVM which did quite well but failed to do well when module had the ability to provide real-time reporting
there is an overlap between classes distribution in the of the abnormal readings and this affirmed the fact
feature space. The analysis of the confusion matrix that the system is a practical one in a constant health
monitoring environment. Overall, the results prove neurophysiological and behavioral data: A multi-
the hypothesis that the offered IoT-ML framework is model, multi-database approach to ADHD
valid, powerful, and effective when applied on the diagnosis", Recent Advancements in Artificial
physiological health measurement in real-time. Intelligence. ICRAAI 2025. Lecture Notes in
Random Forest model happens to be the best reliant Networks and Systems, vol 1468, pp. 429-447,
classifier to be utilized in smart healthcare systems 2025.
run by IoT, which offers to predictive accuracy and [3] N. Mathur, R. Ajmera and J. S. Yadav, "Cause
the stability of the operation. effect testing of flu in medical sciences," IEEE
2017 International Conference on Computing,
4 Conclusion and Future Scope Communication and Automation (ICCCA), pp.
47-52, 2017.
The current paper proposes a Smart IoT-Based Health [4] M. S. Hossain, G. Muhammad, "Cloud-assisted
Monitoring Framework, which can be continuously Industrial Internet of Things (IIoT) – Enabled
energized with the assistance of the Machine framework for health monitoring", Computer
Learning to measure the relevant physiological Networks, Vol. 101, pp. 192-202, 2016.
variables, such as heart rate, oxygen saturation [5] G. K. Soni, D. Yadav, A. Kumar, P. Jain, A.
(SpO2), and body temperature. The model considers Rathi, "Design and SAR Analysis of DGS Based
the measurements of wearable sensors and powerful Deformed Microstrip Antenna for ON/OFF Body
preprocessing algorithms and machine learning Smart Wearable IoT Applications", Physica
classifiers including the Logistic Regression, Support Scripta, Vol. 100, Number 1, pp. 1-28, 2025.
Vector machine (SVM) and the Random Forest to [6] A. Rejeb, K. Rejeb, H. Treiblmaier, A.
identify health conditions as normal or abnormal. The Appolloni, S. Alghamdi, Y. Alhasawi, M.
results of the experiment indicate that the most Iranmanesh, "The Internet of Things (IoT) in
efficient and the predictive accuracy is 95.68 per cent healthcare: Taking stock and moving forward",
and high precision, recall, and F1-score. The IoT Internet of Things, Vol. 22, 2023.
environment will provide the proposed system to [7] C. Costa, J. M. Faria, D. Guimarães, D. Matos,
connect to the cloud that facilitates the smooth flow A. H.J. Moreira, P. Morais, J. L. Vilaça, V.
of data, real-time inferences, and immediate creation Carvalho, "A Wearable Monitoring Device for
of alerts that confirm the feasibility of the system in COVID-19 Biometric Symptoms Detection",
remote patient care. The real-time detection of IRBM, Vol. 84, Issue. 6, 2023.
anomalies with the assistance of IoT, ML, the [8] G. K. Soni, D. Yadav and A. Kumar, "Enhancing
increased robustness of the model through the Healthcare: Flexible and Wearable Antenna
structured pre-processing, and the comparative Design for Tumor Detection," IEEE 2024
analysis of the classifiers to determine the most International Conference on Distributed
beneficial one are some of the significant Computing and Optimization Techniques
contributions. The framework is extensible, (ICDCOT), pp. 1-5, 2024.
dependable, and can be applied in geriatric, chronic [9] G. Gopichand, T. Sarath, A. Dumka, H. R.
disease treatment, and telemedicine. Goyal, R. Singh, A. Gehlot, L. R. Gupta, A. K.
The system can be enhanced by addition of additional Thakur, N. Priyadarshi, B. Twala, "Use of IoT
features such as heart rate, blood pressure, and sensor devices for efficient management of
electrolytes, deep learning algorithms to perform the healthcare systems: a review", Discover Internet
time-related analysis, e.g., CNN and LSTM, and edge of Things, Vol. 4, article No. 8, 2024.
AI to minimize latency and use privacy-preserving [10] B. Pradhan, S. Bhattacharyya, K. Pal, "IoT-
algorithms, e.g., federated learning, to provide Based Applications in Healthcare Devices",
personalized and safe healthcare monitoring. Journal of Healthcare Engineering, 2021.
[11] G. Luo, "PredicT-ML: a tool for automating
References machine learning model building with big
clinical data", Health Information Science and
[1] R. Sinha, "The role and impact of new Systems, Vol. 4, 2016.
technologies on healthcare systems", Discover [12] S. C. Christopoulou, "Machine Learning Models
Health Systems, Vol. 3, Article No. 96, pp. 1-14, and Technologies for Evidence-Based Telehealth
2024. and Smart Care: A Review", BioMedInformatics,
[2] D. Deepika, A. Akanksha, and R. Ajmera, Vol. 4(1), pp. 754-779, 2024.
“Exploring the convergence of