Development of Machine Learning Tool for Rainfall Prediction in Vidarbha Region

Abstract

newline Accurate daily precipitation prediction is important in order to maximize increasing newlineagricultural productivity while preserving food as well as water resources management newlineand also maintaining the health of communities. The study addresses the effectiveness newlineof various ML as well as DL models in forecasting rain, especially in the unexplored newlineVidarbha region, which consists of eleven districts, with Nagpur district being a newlinesuitable example. The research compares conventional ML models linear regression, newlinerandom forest, SVM, as well as XGBoost and sophisticated hybrid models like newlineExtended Short-Term Memory and Integrated Moving Average System with Auto newlineRegression. newlineLinear regression was shown to reduce Root Mean Squared Error and attain very good newlineR-squared, while random forest showed low RMSE and promising R-squared values, newlinethough overfitting was indicated by a negative R-squared. SVM also performed well newlinewith good MSE, MAE, and RMSE values. Among the deep learning methods, LSTM newlineshowed high accuracy with very low RMSE during training and testing times and a newlinehigh R-squared value, ranking it as highly efficient in modeling rainfall trends. newlineARIMA, on the other hand, gave accurate predictions with minimal MSE, MAE, and newlineRMSE. newlineThe study highlights all the potential R2 of both standard and deep learning models for newlineincreasing rainfall forecast accuracy, providing useful information on climate newlineforecasting. These results are important to intelligent options in agriculture as well as newlinewater resource management to enhance better food along with water security in the newlineVidarbha region and other places. newlineThe optimal performance was the one recorded for Buldhana district under the Hybrid newlineARIMA + GRU framework, with at least Mean Squared Error of 0.639553 as well as newlineMean Absolute Error of 0.478201. Gadchiroli district under the ARIMA model saw the newlineworst output by recording high MSE of 144.630877 and an MAE of 5.386492.