Annual Rainfall Prediction of Maharashtra State Using Multiple Regression

Abstract

This paper presents a study of Indian rainfall and prediction of the annual rainfall in the state of Maharashtra and Konkan. The decreasing trends in seasonal rainfall and post-monsoon rainfall and increasing occurrence of the deficit rainfall years indicates the probable intensification of water scarcity in many states and sub divisions of India. Rainfall serves a major source of water only when it is conserved, thus a proper analysis and estimation of rainfall globally is of utmost importance. In an agricultural country like India, where the majority of agriculture is rain dependent rainfall prediction can help to understand the uncertainty in rainfall pattern which may affect the overall agricultural produce. The present study consists a descriptive analysis of annual rainfall in India from 1950–2020, this visualization may prove helpful for deciding the right model for prediction. This study is aimed at finding the most apt model for making accurate prediction for the rainfall dataset. Two machine learning model and a neural network model are implemented and their results are compared. The performance of the results was measured with MSE (mean squared error), RMSE (root mean square error), MAE (mean absolute error). The machine learning models showed high level of deviation as the time series data in use was highly inconsistent, on the other hand the neural network showed better efficiency due to the local dependency in the model which helps it to learn and perform better.