International Journal of Scientific Research in Engineering and Management (IJSREM)

Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

Flood Prediction using Machine Learning

Dr. M. Sengaliappan1, Muthu Sahin S H 2
1
Head of the Department, Department of Computer Applications, Nehru College of Management, Coimbatore,

Tamilnadu, India, ncmdrsengaliappan@[Link]

2
II MCA, Department of Computer Applications, Nehru College of Management, Coimbatore, Tamilnadu, India,

muthusahin123@[Link]

Abstract: Due to urbanization and climate change, flooding has increased in frequency and severity,
upsetting lives and seriously damaging property. Flood Susceptibility Modeling (FSM), which
employs sophisticated machine learning approaches, helps identify flood-prone locations and the
elements that contribute to these risks in order to solve this problem. This study explores hybrid FSM
models that integrate the Index of Entropy (IOE) with Decision Tree (DT), Support Vector Machine
(SVM), and Random Forest (RF) to offer a dependable approach for flood prediction and prevention.
To assess the predictive power and correlations between influencing elements, the study started with
feature selection and multicollinearity analysis. The relationship between several flood-causing
components and their total effect on flooding was measured by IOE. Weighted inputs from these
findings were used to train the hybrid models. Metrics like the Area Under the Curve (AUC) and
other statistical indicators were used to evaluate the models in order to ensure correctness and
reliability. The standalone DT model performed the worst (77.0%), while the hybrid DT-IOE model
had the best prediction accuracy (87.1%), followed by SVM-IOE and RF-IOE. These findings show
that prediction accuracy is increased when machine learning and statistical techniques are combined.
21% of the study region is extremely sensitive to floods, according to the final susceptibility maps,
underscoring the major impact of human-induced factors including land-use changes and urban
growth. By enhancing feature analysis and prediction accuracy, generative AI significantly enhanced
model performance. The significance of hybrid machine learning approaches in developing efficient
flood risk management plans is highlighted by this study, which also supports disaster resilience and
sustainable urban design.

Keywords: Human-induced factors, Flood occurrences, Flood susceptibility modeling

(FSM), and hybrid models Artificial Intelligence (ML), Natural Causes Remote
Observation.

1. Introduction:
One of the most destructive natural disasters, provide precise maps of flood risk and long-
floods have a devastating effect on people, term management strategies that emphasize
infrastructure, agriculture, and the economy. readiness and prevention. Flood prediction is
There is increasing demand on governments to difficult because of the changing climate, and

© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 1

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

physical models limit short-term forecasts by preparedness for catastrophes.

requiring large amounts of data and computer
power. An alternative that is quicker and data-
driven is machine learning (ML), which uses
previous data to efficiently identify flood
patterns with little input. When compared to
conventional techniques, machines (SVMs),
and neuro-fuzzy systems offer superior
accuracy and lower complexity. Combining
machine learning with other methods or
models improves their resilience and Fig 1: Flood Prediction of Each Month
flexibility. But in order to guarantee accuracy
and prevent generalization problems, machine
learning models rely on high-quality data,
necessitation a variety of training datasets. 1.2 GenAI and Deep Learning
How well they function depends on the kind of
prediction (e.g., short-term vs. long-term) and ➢ Artificial neural networks are used in deep
the available data. Despite obstacles, machine learning, a subfield of machine learning,
learning (ML) is still a crucial tool risk which draws inspiration from the structure
management and flood prediction, especially and function of the human brain. Deep
in locations with intelligent sensors or rain learning models, particularly deep neural
gauges. networks, are composed of several layers of
interconnected neurons that can learn
1.1 Flood Prediction Techniques of ML hierarchical data representations. They have
demonstrated great effectiveness in a variety
One of the most damaging natural calamities, of fields, like speech and picture
floods seriously harm both property and human recognition, and are adept at picking up
life. Flood prediction has become increasingly intricate patterns.
important due to climate change induced
increases in rainfall. By learning form past ➢ Generative AI (GenAI) aims to produce
data, machine learning (ML) is essential for fresh content, as was previously said. It falls
forecasting floods and other natural disasters. under the subgroup of Deep Learning. It
Three types of machine learning can be uses generative models, such as generative
distinguished: supervised learning, in which adversarial networks (GANs) and
models are trained on labelled data to predict variational autoencoders (VAEs), to learn
outcomes (e.g., classification for categorical from data and generate new samples with
variables, regression for continuous variables); similar trails. Generative models must first
unsupervised learning, in which models understand the underlying patterns and
identify patterns in unlabeled data (e.g., structures of the data in order to produce
association for identifying relationships outputs that resemble the training data.
between variables, clustering for grouping
similar items); and reinforcement learning. By
aiding in decision-making via the examination
of past data, machine learning (ML) promotes
© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 2
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

2.1 Working of Modules

Using data validation, the prediction accuracy
of various models is assessed, and the accuracy
is obtained by comparing the results. By
comparing algorithms, the false-positive rate
specification precision, recall, training dataset
Fig 2: Understanding with GenAI and Deep accuracy, and testing dataset correctness are
Learning determined.
The steps involved:
2. Proposed Method: • Define a problem
The Indian Meteorological Department’s • Preparing data
rainfall data for Kerala (1901-2018), which • Evaluating algorithms
used machine learning methods like KNN, LR, • Prediction result
SVM, DT, and RF to analyze the millimeter-
scale data. Performance evaluation and quality
indicator (SNIP, Cite Score, SJR, h-index)
were used to rank peer-reviewed studies on
flood prediction using machine learning. In
order to determine which models performed
best for particular application kinds, and
results. Applications for flood prediction
depend on important factors such as
streamflow, rainfall, water level, river flow,
and soil moisture. Rainfall has a major impact Fig 3: System Architecture
on flood modelling and runoff, particularly for
short-term forecasts and flash floods. Rainfall
by itself, however, is not enough to accurately 2.2 Rainfall Data Set
anticipate floods, especially in long-term A dataset that includes historical flood data and
scenarios when catchment conditions and soil rainfall information for particular regions, such
moisture are crucial. The most important flood as Kerala, is employed. With average rainfall
resource variable is covered in this paper, along estimated every ten days and displayed on a
with how they can be integrated with machine graph, the dataset covers around three months.
learning (ML) techniques. It highlights how Flood occurrences are employed as output
various machine learning approaches depend labels in a machine learning model trained on
on the dataset, use case, and kind of forecast Kerala’s yearly rainfall data. The model can be
(e.g., short-term water level projections or used to predict floods in any Indian state with
long-term streamflow modeling). comparable data because it is trained and
stored using daily rainfall criteria.

© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 3

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

column is added based on meteorological

data, and labels are transformed into
numericrepresentations for machine
readability. The process entails loading the
dataset, handling null values effectively,
importing libraries like Numpy and
Pandas, and saving the cleaned data as a
CSV file for further use.
Fig 5: Data Pre-Processing

Fig 4: Rainfall Data Set

2.1.3 Model Selection and Prediction

2.1.1 Data Preparation
This study aims to collect and organize rainfall
data from Kerala and other regions of India in
order to investigate flood-prone areas. Data
needs to be collected, aggregated, profiled,
verified, and converted before it can be used
for analytics and visualization. It entails
gathering data from internal and external
sources and incorporating it into warehouses,
data lakes, and NoSQL databases. This
procedure, which is frequently referred to as
“data prep” or “data wrangling,” is crucial for
creating analytics applications and is
completed by analysts, data scientists, and IT
teams via self-service technologies.

2.1.2 Data Pre-Processing

Fig 6: ML Model Selection Process
The first stage in getting raw data ready for
machine learning models is data pre-
processing, which makes sure the data is clean Selecting the optimal model for a job is known
and formatted. This procedure eliminates as model selection. Floods are predicted using
missing, null, or duplicate values that are methods such as SVM, Random Forest, KNN,
frequently present in raw datasets. A “flood” Decision Tree, and Linear Regression. These
© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 4
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

models examine input data, spot trends, and outputs. Effective algorithms for classifying
forecast outcomes. To Find patterns and learn, rainfall data include SVM, Random Forest,
they are trained on datasets. The decision is KNN, LR, and Decision Tree.
influenced by variables such as the complexity,
characteristics, and quantity of the dataset. It is
advised to begin with a basic model, Classification Task Types:
progressively add complexity, and use cross- • One of two classes is predicted by
validation and parameter adjustment to binary classification, such as spam
maximize accuracy. or non-spam.
• Predicting one of more than two
classes (such as flower types) is
2.1.4 Splitting Dataset into Train and Test known as multi-class classification.
Data • Several classes are predicted for
every instance using multi-label
Data pre-processing involves splitting the
classification (e.g., tagging
dataset into training and test sets in order to
photographs with several labels).
improve model performance in machine
• Imbalanced Classification:
learning.
Addresses unequal distribution of
Training Set: A subset of the dataset with classes (fraud detection, for
known results that is used to train the model. example).
Test Set: A subset of the dataset used to access 2.1.6 Tasks including email filtering, fraud
detection, medical diagnosis, and weather
the model’s forecasts. forecasting all make extensive use of
classification. Certain problem types are
Action to take: Divide the dataset into tests and
better suited for different algorithms, and
train subsets at random. in order to attain the best results, model
parameters must be how well the data is
• X_train: Training features classified, hence data pre-processing is
• X_test: Test-related feature essential. Prediction Result
• Y_train: Training dependent variables Generation
• Y_test: Testing dependent variables
To guarantee consistent results, use
train_test_split() with: Data arrays test_size
for train-test ratio random_state.

2.1.5 Classification
A supervised learning method for classifying
fresh observations using training data is the
classification algorithm. With classes standing in
for objectives or labels, it learns from labelled
datasets and divides data into categories like Fig 7: Prediction Process of Flood
Yes/No or 0/1. Finding the input data’s category
is the main objective, especially for categorical
© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 5
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

LASS 1: “FLAH FLOOD MAY power, importance, and correlations between

OCCUR”: these variables—which were found to have a
This class is set to 1 if the input is greater than major influence on flood occurrences—were
2400mm. examined using feature engineering. The
dearth of earlier research on the region’s
CLASS 2: “FLASH FLOOD MAY NOT
vulnerability to flooding is addressed in this
OCCUR”: The flash flood class is set to 0 it the
paper.
input is less than 2400mm.

3.1 The actions listed below demonstrated that

2.3 Materials and Methods
the suggested model offers a very simple and
The dataset used for the analysis was rainfall effective way to forecast floods:
data from 1901 to 2018. The dataset is arranged
by month, state, and district and is generated as
Step 1: Pre-processing is done on the rainfall
a CSV file. The unit used to measure rainfall is
the millimeter (mm). The dataset was collected collection dataset.
monthly from 36 locations in the metrological Step 2: The rainfall dataset is divided into
department. By examining historical data, training and testing groups at random.
such as flood events in the past, machine
learning technology can be used to forecast the Step 3: LR, DT, KNN, RF, and MLP algorithms
future. LR, SVM, KNN, and MLP are some of were used to train the dataset.
the methods used to measure ML performance. Step 4: The most accurate SVM method is used
to build the model, and it is verified using
3. Result: metrics like accuracy, sensitivity, specificity,

Using machine learning techniques, the recall, precision, and f1-score.

proposed work assesses a rainfall dataset to Step 5: Feed the prediction model test data and
make highly accurate prediction about flash confirm the outcomes. The algorithm's
flood. Using training and testing datasets, flood
calculated accuracy.
models' performance is assessed using ROC
curves and statistical markers. The study’s Step 6: The flood warning system also
main goals are to create maps of flood-prone appears.
areas and construct machine-learning-based
flood susceptibility models by examining a
variety of influencing factors, such as natural
and of human origin causes. The predictive

Fig 8: Prediction Label and

Accuracy
© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 6
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

3.2 Performance Analysis of Algorithm: Classification Algorithm

Discussion
The procedures listed below, along with the
suggested model, offer a relatively simple By examining a variety of parameters in the
study area, this research paper aims to create
and effective way to forecast flooding:
machine learning-based flood susceptibility
models that will provide maps that are prone
to flooding. Both natural and of human origin
Step 1: Pre-processing is done on the rainfall
element that have been show to affect flood
collection dataset. events were taken into account. The
Step 2: The rainfall dataset is divided into predictive power, importance, and
correlations between these characteristics
training and testing groups at random.
were evaluated through feature engineering,
Step 3: The dataset was trained using the LR, and the results were utilized for modelling.
SVM, KNN, RF, and DT algorithms in step Finding and assessing region-specific
three. characteristics was crucial because there is no
previous research in this field. Determining
Step 4: The model is built with the highest contributing elements, confirming previous
accuracy using the SVM method and verified flood disasters, and identifying regions that
using characteristics like accuracy, precision, are prone to hazard were all made possible by
remote sensing. Effective machine learning
confusion matrix, and fi-score.
methods including DT, SVM, and RF were
Step 5: Feed the prediction model test data used both alone and in conjunction with IOE
and confirm the outcomes. to create accurate flood susceptibility maps.
The goal of the project is to assess rainfall
Accuracy of the algorithm calculated from data in order to make accurate flash flood
the F1 score and Confusion Matrix. forecasts.
From (1) and (2), the precision measured.

4. Conclusion
In addition to encouraging collaboration
between public institutions for better flood
prediction and warning system, this initiative
offers insights into the needs and readiness of
local communities. By connecting
stakeholders, public officials, and citizens, a
cooperative platform will enable prompt
flood warnings to lessen negative effects.
Flash floods seriously harm people and
property. A model was created to predict flash
floods using rainfall data from 1901 to 2018.
The pre-processed dataset was split between
Fig 9: Performance Analysis of
80% training and 20% testing, and it was
© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 7
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

analyzed using the SVM, Logistic Regression, learning,” J. Hydrol., vol. 585, no.
KNN, Decision Tree, and Random Forest December 2019, p. 124808, 2020, doi:
algorithms. Calculations 10.1016/[Link].2020.124808.
were made for performance criteria such as
sensitivity, specificity, F1 score, recall, and I. E. Olorunfemi, A. A. Komolafe, J. T.
accuracy. SVM successfully forecasts floods Fasinmirin, A. A. Olufayo, and S. O.
based on rainfall data and reached the best Akande, “A GISbased assessment of the
classification accuracy, surpassing 90%. potential soil erosion and flood hazard
Disaster management agencies can use this zones in Ekiti State, Southwestern
model to help predict floods. The goal of Nigeria using integrated RUSLE and
future research is to automate prediction HAND models,” Catena, vol. 194, no.
outputs using desktop or online applications January, p. 104725, 2020, doi:
and integrate cutting-edge AI approaches. 10.1016/[Link].2020.104725.

5. P. T. Padi, G. Di Baldassarre, and A.

Reference: Castellarin, “Floodplain management in
Africa: Large scale analysis of flood
1. Y. Wang, Z. fang, H. Hong, and L. Peng,
data,” Phys. Chem. Earth, vol. 36, no. 7–
“Flood susceptibility mapping using
8, pp. 292– 298, 2011, doi:
convolutional neural network
10.1016/[Link].2011.02.002.
frameworks,” J. Hydrol., vol. 582, no.
March, p. 124482, 2020, doi:
6. I. Ajibade, G. McBean, and R. Bezner-
10.1016/[Link].2019.124482.
Kerr, “Urban flooding in Lagos, Nigeria:
Patterns of vulnerability and resilience
2. R. Mind’je et al., “Flood susceptibility
among women,” Glob. Environ. Chang.,
modeling and hazard perception in
vol. 23, no. 6, pp. 1714–1725, 2013, doi:
Rwanda,” Int. J. Disaster Risk Reduct.,
10.1016/[Link].2013.08.009
vol. 38, no. April 2018, p. 101211, 2019,
doi: 10.1016/[Link].2019.101211.
7. J. Ntajal, B. L. Lamptey, I. B.
Mahamadou, and B. K. Nyarko, “Flood
disaster risk mapping in the Lower Mono
3. W. Chen et al., “Modeling flood
River Basin in Togo, West Africa,” Int. J.
susceptibility using data-driven
Disaster Risk Reduct., vol. 23, no.
approaches of naïve Bayes tree,
October 2016, pp. 93–103, 2017, doi:
alternating decision tree, and random
10.1016/[Link].2017.03.015.
forest methods,” Sci. Total Environ., vol.
701, 2020, doi:
8. I. Douglas, “Flooding in African cities,
10.1016/[Link].2019.134979.
scales of causes, teleconnections, risks,
vulnerability and impacts,” Int. J.
4. R. Costache et al., “Spatial predicting of
Disaster Risk Reduct., vol. 26, no.
flood potential areas using novel
September, pp. 34–42, 2017, doi:
hybridizations of fuzzy decision-making,
10.1016/[Link].2017.09.024
bivariate statistics, and machine

© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 8

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 09 Issue: 02 | Feb - 2025 SJIF Rating: 8.448 ISSN: 2582-3930

9. C. C. Olanrewaju, M. Chitakira, O. A.
Olanrewaju, and E. Louw, “Impacts of
flood disasters in Nigeria: A critical
evaluation of health implications and
management,” 52 Jamba J. Disaster Risk
Stud., vol. 11, no. 1, pp. 1–9, 2019, doi:
10.4102/jamba.v11i1.557.

© 2025, IJSREM | [Link] DOI: 10.55041/IJSREM41892 | Page 9