PREDICTING RAINFALL BASED ON MACHINE

LEARNING MODELS
Ms. M.D.Niveedha Ms. M.Deepa Ms. G.Harika
UG Scholar , Department of CA, UG Scholar , Department of CA, UG Scholar , Department of CA,
Mohan Babu University Mohan Babu University Mohan Babu University
(Erstwhile Sree Vidyanikethan (Erstwhile Sree Vidyanikethan (Erstwhile Sree Vidyanikethan
Engineering College(Autonomous), Engineering College(Autonomous), Engineering College(Autonomous),
Tirupati, Andhra Pradesh,India Tirupati, Andhra Pradesh,India Tirupati, Andhra Pradesh,India
[email protected] [email protected] [email protected]

Ms. G.Rishitha Reddy

UG Scholar , Department of CA,
Mohan Babu University
(Erstwhile Sree Vidyanikethan
Engineering College(Autonomous),
Tirupati, Andhra Pradesh,India
[email protected]

ABSTRACT Keywords: Machine learning, Rainfall, support

Adequate rainfall estimation bears significant relevance vector machine, Random forest, logistic regression,
towards enhancing the resilience of disaster mitigation weather forecasting.
measures as well as the impacts of high risk climatic
occurrences including floods and droughts. Standard
1 .INTRODUCTION
observational methods when practiced often provide a
The ability to forecast the amount of precipitation is
systematic way of measuring weather parameters but
an essential aspect of climate science and
often they do not provide the resolution or precision that
adequately measures the prevailing intensity and duration considerably impacts ecosystem management,
of rainfall during storms. Therefore, there is a need to agriculture, urbanization, and disaster management.
apply sophisticated procedures in order to enhance the Rainfall prediction is essential for the management of
accuracy of such predictions. In terms of applicability, water resources at [1] Analyzing significant amounts
parameters of ML algorithms are considered appropriate of data sets using complex patterns and relationships
to apply, primarily due to its mitigated capability of has made it possible to employ effective machine
identifying nonlinear and other high-level interactions to learning (ML) techniques in rainfall forecasting. This
gain nearly accurate predictions with high intensity
research investigates the potential of ML models
under unfavorable circumstances.. The proposed system
applies a machine learning integration technique where
applied on four different time frames, which include
the other can be incorporated into the system operation daily, weekly, 10 day and monthly time frames as
for better accuracy and speed of the forecast. For well as showing their precision level in rainfall
machine learning support vector machines, random predictions. Through sophisticated modeling
forests and logistic regression are provided. Since SVM approaches and normalization techniques, the
can easily handle high dimensionality, random forest research shows how these methods are capable and
which is very stable and fight against overfitting lastly, accurate in predicting the rainfall pattern. It uses
the logistic regression which gives the chance of these advanced forecasting techniques to generate
occurrence of an event Whole forecasting model.
practical recommendations for the management of
Initially, the whole data is cleaned through data cleaning,
normalized through normalization and finally featurized water reservoirs and the coping strategies to climate
through feature engineering step. This study reaffirms variability.
the critical importance of this technological advance to fit The ability to accurately predict rainfall is pivotal for
closely with institutional frameworks and crisis the management of water resources, disaster
management systems. planning, agricultural activities, particularly in areas
affected by erratic climate changes.[2] In dry areas
like the united arab emirates uae, there are serious
problems caused by low water availability and erratic
climate conditions. The prediction of rainfall stress
seems to be useful in mitigating these problems and
managing water resources. The present research uses
advanced artificial intelligence techniques to boost
the rainfall predictions on a monthly basis in hyper
arid environments using a robust thirty year data set
from 1991 to 2020. These predictions help to
understand these patterns in arid regions.
Based on the information of the Niavaran station
located in Teheran, In [3] The forecasting of rainfall
from an hour to a few days is very vital for the
achievements of hydrological modeling and other
aspects such as urban flood warning systems and Fig1: predicting rainfall using machine learning
urban water drainage management. This paper seeks model
to model rainfall for relatively short lead times of 5 The importance of RF is derived from its ability to
and 15 minutes from a station in Tehran, Iran. The perform constant and accurate calculations
models were further improved by considering the independent of moderately high-dimensional data.
intensity and duration of rainfall as factors resulting The study investigates rainfall prediction models
in a better model accuracy. using machine learning algorithms such as Random
Substituting traditional weather models with ML Forest (RF), and Support Vector Machines (SVM).
models, as applied to predicting rainfall amounts, In [5],This improved prediction capacity highlights
yields a number of benefits. Machine learning the requirement for implementing advanced
algorithms that include a number of variables such as methodologies to resolve problems associated with
temperature, humidity, wind speed, and atmospheric time-dependent data. Better predictive performance
pressure, are able to learn from such high- helps make decisions for agricultural purposes, and,
dimensional data and provide outputs with low errors simultaneously, assists in general attempts at building
invariably. Such models also differ in how they are climate resilience, disaster preparedness, and more
developed where traditional models are often based sustainable use of water resources. This research
on a linear specification while machine learning brings the possibility of technological advance
models are more flexible in integrating into practice changing traditional practices and providing robust
tools in one of the most critical challenges within
the underlying nonlinearities of the data. In addition,
environmental and agricultural planning to the
these models have the ability to periodically learn
forefront, given its focus on models able to harness
and enhance their performance over time as addition
temporal dynamics. The following Figure 2 depicts
of fresh data, this aids them in being resilient in a
the comparison of different machine learning models
dynamic environment..The following figure 1 decipts
rainfall prediction using machine learning model.
Rainfall estimation with the use of machine learning
architectures like the Random forest (RF), Support
vector regression (SVR) and In[4].A new trend in
rainfall estimation has been the application of
interferential machine learning models, which deliver
better accuracy and flexibility. Of these, Random
Forest (RF) is one of the most stable algorithms that
can contribute to decreasing the data overfitting and
handling large and loading datasets. RF works by
building approximately hundreds of decision trees, Fig2: comparison of different machine learning
combined with the final decision, thus it is not easily models
affected by noise and can give accurate results even
The aim of this study is to evaluate the
when tested in different climate conditions.
effectiveness of forecasting rainfall using machine
learning models like support vector machines supervised algorithms: Decision Tree, NFiguraïve
( SVM), artificial neural networks (ANN), random Bayes, K-Nearest Neighbors, and Support Vector
forests (RF).[6] The objective of this work is to Machines, using fuzzy logic to increase the
compare the capability of several machine learning- prediction accuracy.. With the help of weather data
from Lahore spanning 12 years (from 2005 to 2017),
based forecasting techniques for rainfall against
it was established that the fusion-based framework
more conventional statistical methods. Data models
made it possible for maximal rainfall prediction
that have the best ability to capture complex
performance The SVM outperformed other models
relations in the data, both in nonlinear relations and and presented the highest efficiency, being strong
even in data with inconsistencies are explored in the with respect to tackling complex high-dimensional
research. Among these approaches, one stands out data sets. SVMs have the advantage of being capable
for the strength in learning intricate interactions in of both linear and nonlinear classifications and
data and robustness to noisy or missing data. resistant to over fitting in high dimensions. However,
Robustness is an attractive property for they are very computationally intensive and their
meteorological practice because the accuracy and performance depends largely on the choice of their
variability of data are serious sources of challenge. parameters . The framework incorporated an
This means that the sophisticated machine learning adaptive mechanism that could adapt to variability
methodology has the potential to provide more and unpredictability in weather data to make it more
accurate prediction results that may not be produced robust and efficient. The authors suggest using
through the traditional approach because such an multiple algorithms which best complement one
approach usually neglects the non-linear and another to optimize the results obtained with SVM as
the most effective in processing the
dynamic character of rainfall data. Such findings
meteorological dataset.
reveal the possibility that could be seen when
embracing modern, advanced predictive tools and, Rainfall prediction is crucial in meteorological
thus, enhancing the precision of predictions for applications, particularly in areas with variable
sectors such as water supply management and climatic conditions In 2022[8] Nana Kofi Ahoi
agricultural planning and broader climate resilience Appiah-Badu et al. investigated five machine
programs. learning algorithms (Decision Tree, Random Forest,
Multilayer Perceptron, XGBoost, and K-Nearest
Using machine learning methods in rainfall Neighbor) using data from 1980 to 2019 from the
prediction enhances the accuracy of the forecasts. Ghana Meteorological Agency.. Random Forest and
This is an improvement that can help communities XGB significantly outperformed their rivals,
in their ability to adapt to climate change, manage achieving great success in both rain and no-rain
their water resources properly, and minimize classifications, whereas MLP exhibited fairly high
hazards associated with storm events. So machine accuracy but rather longer execution times, and KNN
learning algorithms help to predict the rainfall with performed worst. as its predictions rely on proximity,
the large datasets and suitable algorithms. Some of and thus the performance suffers significantly if the
dataset size is not sufficiently large. The Decision
the techniques we use to predict rainfall are SVM,
Tree algorithm is not very accurate as that of RF or
random forest, and also logistic regression in this
XGBoost but executes the fastest; hence, it can be
study area And are applied over the large datasets
applied for real-time prediction, especially if speed is
and thus performance are measured and predictions essential. The study points out the trade-offs between
are done. accuracy, computational efficiency, and speed in the
2. LITERATURE SURVEY selection of machine learning algorithms for
meteorological applications.
This Section provides an extensive Literature Survey
on Existing Mechanisms for Predicting Rainfall using In the year 2023 the study conducted by Md. Mehedi
Machine Learning Models. In the year 2021 [7] Hassan[9]. Explored machine learning algorithms for
Chalachew Muluken Liyew and Haileyesus Amsaya rainfall prediction using Australian weather data,
Melese to improve real-time rainfall prediction for including Naïve Bayes (NB), Decision Tree (DT),
smart cities. Proposed framework integrated four Support Vector Machine (SVM), Random Forest
(RF), Logistic Regression (LR), and Artificial Neural was performed on DNN and LSTM models, used to
Networks (ANN). Techniques like Principal capture nonlinearity and sequence structure in data. A
Component Analysis (PCA) and k-means clustering comparative analysis in the current study showed that
were employed to enhance model performance and even though SVM and KNN are useful when making
analyze regional rainfall patterns. The ANN achieved classification-based forecasts, they could not
the highest overall accuracy, 90% before feature adequately consider temporal features necessary to
selection and 91% after. Whereas the RF and SVM predict rainfall accurately. However, DNN and
potentially offered comparable results, following the LSTM provide more detailed analysis by utilizing
ANN. This shows the strength of those algorithms, deep learned structures, yet do it less efficiently. This
especially capturing non-linear and dynamic duality of simplicity, exactness, and complexity is
relationships, which are inherent in meteorological extended caution to choosing algorithms depending
data. This, therefore, stresses that such machine on the needs of the forecast task in Ning.Yu and
learning models ought to be integrated with Haskins’ work does a good job in adding to the body
appropriate preprocessing techniques for data of knowledge by developing a framework that
improvement purposes to enhance the accuracy of integrates both classical and contemporary ML
forecasting. The research helped enhance the approaches for enhancing systems for rainfall
interpretability of results through PCA and clustering predicting. They particularly stress the need to
methods and provided information about the regional choose and fine-tune the algorithm as a key to
variation in rainfall. This approach will focus on the improving the quality of both the forecast and the
applicability of machine learning as a robust tool to underlying process in meteorological applications.
address certain issues in meteorology. This has larger
In [11] the study conducted by Martin Kuradusenge,
implications on the management of water resources,
of this study is in the use of integration of
agricultural planning, and disaster preparedness.
meteorological data from several states to enrich the
These new results are the contribution toward a still-
accuracy of rainfall predictions for a specified target
ongoing trend in the literature about urging to bring
region, that is Delhi, India. By combining diverse
advanced computational methods toward applications
climatic data from Uttarakhand, Haryana, Punjab,
in environmental science. From the opposite view,
Uttar Pradesh, Himachal Pradesh, Madhya Pradesh
these results point out many exciting futures works
and Rajasthan States, the study leads to a pioneering
that could be further analyzed, such as the hybrid
cross-regional data-integration approach The
models and development of ensemble approach and
application of machine learning (ML) techniques,
thus possibly increase predictability accuracy through
particularly Random Forests (RF) and Logistic
resolution of the restrictions specific for pure,
Regression (LR), has been widely studied for rainfall
standalone algorithms.
prediction due to their ability to model complex, non-
In [10] Ning Yu, and Timothy Haskins, created a linear relationships with environmental data. RF is an
framework that combines classical and advanced ensemble decision tree method that is effective for
machine learning algorithms for rainfall forecasting handling large datasets; however, it can suffer from
in Upstate New York. It evaluated various models, overfitting, especially when faced with noisy data.
namely K-Nearest Neighbors (KNN), Support Vector Conversely, LR has demonstrated better performance
Machine (SVM), Deep Neural Networks (DNN), and in accuracy, particularly when antecedent cumulative
Long Short-Term Memory (LSTM). Results rainfall data is included. Research indicates that the
indicated that. SVM turned out efficient at the inclusion of various atmospheric features, like
classification level owing to its ability to work on temperature, humidity, wind speed, and pressure, can
hyperplanes which can help detect the rainfall significantly improve rainfall prediction accuracy
pattern. In addition to this, it is sturdy and capable of using ML techniques. When considering accuracy in
providing good solutions even with noisy inputs in prediction LR consistently outperforms RF,
large numbers of features. In the same regard, simple suggesting it may be a more reliable model for
and popular classifier, KNN did well when it came to rainfall prediction and warning systems. In essence,
classification. However, due to the prediction time LR stands out as a precise and efficient method for
being higher because of the distances algorithm for predicting rainfall.
every instance, this model cannot be used effectively
3. METHODOLOGY
in real-time forecast scenarios. Another experiment
The following key components are described in this
section: data collection, Data preprocessing, Model 3.3.2. LOGISTIC REGRESSION
selection, Evaluation metrics. Logistic regression is one of the statistical method
capable of modeling the odds of an observation
3.1 DATA COLLECTION belonging to either of the two classes, used. The
Said data collection is a critical process,these datasets model is built based on the logistic (sigmoid)
contain parameters such as pressure, temperature, equations
maximal pressure, humidity, wind speed, direction,
dew point. The data is presented in tabular format. 3.3.3. RANDOM FOREST
This Random Forest is a stable and flexible machine
3.2 DATA PREPROCESSING learning method that has been widely used in the
Data preprocessing might be largely considered as analysis of rainfall data because of its effectiveness in
being a crucial stage, The dataset underwent some handling dynamic data matrix and generalizing high
pre-processing to deal with issues of missing values, prediction accuracy.
wrong encoding and redundant data. The following figure 5 depicts the architecture of
The textual representation of the weather was random forest
preprocessed from the categorical format to
numerical. The following Figure 3 depicts the
process of machine learning techniques

Fig5: Architecture of Random forest

3.4. EVALUATION METRICS

Fig 3: process of machine learning technique The motive of this study is to evaluate the
performance of the machine learning algorithms
3.3 MODEL SELECTION 3.4.1. PRECISION
The following research seeks to determine the factors Accuracy is one of the objectives of the evaluation
that facilitate the use of machine learning techniques matrix where it is applied to measure the values of
in predicting Rainfall the positive prediction.
Precision=TP/ TP+FP
3.3.1. SUPPORT VECTOR MACHINE
SVM is one of the most effective supervised learning 3.4.2. RECALL
methods that is used widely in classification. The Recall previously known as sensitivity, identifies the
main goal of SVM is to have as much space from proportion of the actual positive cases detected by the
each other as possible all the different classes of data model. It is represented as
with the condition that the hyperplane must be Recall=TP /TP+FN
located at maximum distance from the two closest
points belonging to different classes. 3.4.3. F SCORE
The following figure 4 depicts the architecture of The F1-Score is an assessment that combines both
support vector machine precision and recall, and is useful when using a single
measure to evaluate the performance of a model
. F-score=2*(Precision * Recall)/Precision+ Recall

3.4.4 SUPPORT
The support gives the number of times each class
actually appears in a dataset.

Fig 4: Architecture of support vector machine 4. CONCLUSION

Environmental factors are very much involved in models for gap filling daily rainfall series in
assessing rainfall forecasting models assembling a semiarid region of Spain." Atmosphere
rainfall datasets of environmental factors such as 12.9 (2021): 1158.
rainfall, temperature, humidity, wind direction and
prevailing season. Different fields require knowledge 5. Poornima, S., et al. "Rainfall forecast and
and control of rainfall impacts. This means that the drought analysis for recent and forthcoming
sensor and meteorological datasets should years in India." Water 15.3 (2023): 592.
incorporate more environmental characteristics to
increase accuracy of the rainfall prediction. 6. Mishra, Pradeep, et al. "Modeling and
Rainfall prediction is very significant to a number of forecasting rainfall patterns in India: a time
very important areas like flood forecasting, series analysis with XGBoost algorithm."
agriculture, health and water sharing. In this research Environmental Earth Sciences 83.6 (2024):
paper, the Algorithms used are Random forest (RF), 163.
Support vector machines (SVM ), logistic regression
(LR) machine learning algorithm which is used to 7. Liyew, Chalachew Muluken, and Haileyesus
consider rainfall data In addition to the other external Amsaya Melese. "Machine learning
& internal factors and tried to develop a model to techniques to predict daily rainfall amount."
prediction rainfall. In conclusion Random Forest Journal of Big Data 8 (2021): 1-11.
algorithm’s performance was higher in accessing the
dataset and provided rainfall Prediction with more 8. Appiah-Badu, Nana Kofi Ahoi, et al.
accuracy than the other algorithms. "Rainfall prediction using machine learning
algorithms for the various ecological zones
REFERENCES of Ghana." IEEE Access 10 (2021): 5069-
1. Ridwan, Wanie M., et al. "Rainfall forecasting 5082.
model using machine learning methods:
Case study Terengganu, Malaysia." Ain 9. Hassan, Md Mehedi, et al. "Machine
Shams Engineering Journal 12.2 (2021): learning-based rainfall prediction: Unveiling
1651-1663. insights and forecasting for improved
preparedness." IEEE Access 11 (2023):
2. Baig, Faisal, et al. "How accurate are the 132196-132222.
machine learning models in improving
monthly rainfall 10. Yu, Ning, and Timothy Haskins. "Bagging
machine learning algorithms: A generic
prediction in hyper arid environment?." computing framework based on machine-
Journal of Hydrology 633 (2024): 131040 learning methods for regional rainfall
forecasting in upstate New York."
3. Aderyani, Fatemeh Rezaei, S. Jamshid Informatics. Vol. 8. No. 3. MDPI, 2021
Mousavi, and Fatemeh Jafari. "Short-term
rainfall forecasting using machine learning- 11. Kuradusenge, Martin, Santhi Kumaran, and
based approaches of PSO-SVR, LSTM and Marco Zennaro. "Rainfall-induced landslide
CNN." Journal of Hydrology 614 (2022): prediction using machine learning models:
128463. The case of Ngororero District, Rwanda."
International journal of environmental
4. Bellido-Jiménez, Juan Antonio, Javier research and public health 17.11 (2020):
Estévez Gualda, and Amanda Penélope 4147
García-Marín. "Assessing machine learning