Rainfall prediction using artificial neural networks and machine learning algorithms over the Coimbatore region

ABSTRACT

Due to ongoing climate change, accurately predicting rainfall has become increasingly critical. This paper explores an approach utilizing two different machine learning algorithms, including multilayer perceptron neural networks (MPNN) and random forest regressors (RFR), to enhance rainfall forecast accuracy. Historical daily weather data spanning 100 years (1913–2023) from the Agro Climate Research Centre at Tamil Nadu Agricultural University were used. The study focused on global climate drivers like the Southwest Monsoon (SWM) and Northeast Monsoon (NEM) over the Coimbatore region; this region receives more rainfall during NEM. Normalization and scaling techniques addressed missing values, preserving 70–85% of the original data for the training set. Results demonstrated that MPNN outperformed RFR, achieving an accuracy of 85.55% for SWM and NEM, while RFR outperformed MPNN, producing an accuracy of 86.50%. The coefficient of determination (R2) for predicted versus observed values was 0.8 for daily rainfall from 2020 to 2023.