An Offline Rainfall Prediction Using MLR and Ensemble Learning

Abstract:Accurate rainfall prediction is vital for various sectors in Ghana, including agriculture, hydrology, anddisastermanagement. In this study, various regressor models like Multiple Linear Regression (MLR), Gradient Boosting(GB), Extreme Gradient Boost (XGB), and a Voting Regressor ensemble approach are included. The dataset, consisting of variousclimatic attributes, was sourced from NASA Power spanning 1982 – 2023. The Mean Squared Error (MSE) and explainedvariance were used as evaluation metrics. The default splitting ratio for training and testing data included is 75:25, whichisapplied to all the machine learning models. Our results indicate that the advanced ensemble methods, Gradient Boosting, Extreme Gradient Boosting, and Voting Regressor, outperform the traditional Multiple Linear Regression model interms ofboth MSE and explained variance. These models exhibit superior predictive capabilities, capturing nonlinear relationships andinteractions among predictor variables more effectively. Furthermore, the ensemble approach, by combining the strengths ofindividual models, enhances the overall predictive performance, resulting in a lower error rate for the target location. Ourresearch highlights the importance of utilizing advanced regression techniques, as evidenced by a mean squared error (MSE) score of 11.56.  Keywords: Multiple Linear Regression, Gradient Boosting, Extreme Gradient Boosting, Voting Regressor, MSE, Machine Learning