Machine learning algorithms for predicting rainfall in India

Due to the changing climate and frequent occurrence of extreme events, farmers face significant challenges. Precise rainfall prediction is necessary for proper crop planning. The presence of nonlinearity and chaotic structure in the historical rainfall series distorts the performances of the usual prediction models. In the present study, algorithms based on complete ensemble empirical mode decomposition with adaptive noise combined with stochastic models like autoregressive integrated moving average and generalized autoregressive conditional heteroscedasticity; machine learning techniques like random forest, artificial neural network, support vector regression and kernel ridge regression (KRR) have been proposed for predicting rainfall series. KRR has been considered to combine predicted intrinsic mode functions and residuals generated by various algorithms to capture the volatility in the series. The proposed algorithms have been applied for predicting rainfall in three selected subdivisions of India, namely, Assam and Meghalaya, Konkan and Goa, and Punjab. An empirical comparison of the proposed algorithms with the existing models revealed that the developed models have outperformed the latter.