Predicting Daily Rainfall Amount Using Random Forest, Multivariate Linear Regression, and Extreme Gradient Boosting

To utilize water resources optimally, plan agricultural activities properly, and manage the risks of natural disasters in a country such as India where the economy relies heavily on the monsoon rains, accurate rainfall prediction on a day-to-day basis becomes very crucial. This work considers rainfall prediction of several regions in India and conduct a comparative evaluation of three machine learning models - Multivariate Linear Regression, Random Forest, and Extreme Gradient Boosting. The models were built and validated based on a rich database that contained meteorological variables that correlate with rainfall. The effectiveness of these models was assessed concerning Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). It is evident from our results that the predictive capabilities of any of the models are good, however, the XGBoost model seems to shine out as the best since it had the least RMSE and MAE values even when compared to Random Forest and Multivariate Linear Regression that were quite close. This performance hierarchy indicates that such performance of XGBoost may be a result of its potential of modeling complex non-linear relationships among the variables making it appropriate for the intricate task of predicting rainfall. The results show the prospects for further improvement in rainfall prediction with the application of more sophisticated machine learning approaches, which can noticeably integrate the complex relations between the obtained data and impacts on rainfall.