ADVANCES IN MACHINE LEARNING-BASED RAINFALL FORECASTING: A SYSTEMATIC REVIEW

Abstract

Rainfall forecasting plays a critical role in managing water resources, reducing the risks of extreme weather, and supporting agriculture. Traditional methods such as Numerical Weather Prediction (NWP) and statistical models often struggle with accuracy and scalability, especially when dealing with localized weather conditions or extreme events. In recent years, machine learning (ML) has emerged as a powerful tool for identifying complex patterns in meteorological data. This review explores the progress of ML-based rainfall forecasting from 2014 to 2024. It organizes the literature into five main groups: Traditional Machine Learning models, General Neural Networks, Deep Learning models like LSTM and CNN, Ensemble techniques, and Hybrid models. The paper also highlights key trends through an N-gram analysis of publications from Scopus and Web of Science. The main contributions of this review include an overview of model development over the last decade, a comparison of model performance across different forecasting scenarios, and a summary of which models are best suited for various data environments. While ML models show strong potential, they also face ongoing challenges such as overfitting, high computational costs, and limited interpretability. This review concludes by identifying future research directions to improve the efficiency, transparency, and real-world applicability of ML-based rainfall forecasting.