Different Time-Increment Rainfall Prediction Models: a Machine Learning Approach Using Various Input Scenarios

This study investigates the utilization of machine learning techniques, including Linear Regression, Gradient Boost, and LSTM algorithms, for rainfall prediction across different timeframes (hourly, daily, and monthly). Data spanning from 2015 to 2022 from meteorological stations in the Langat basin river region (Pejabat, Kajang, and Petaling) is employed for model development and evaluation. The primary objectives encompass crafting predictive models, assessing their ability to capture rainfall patterns, and analyzing the impact of various input parameters on model performance. Emphasizing the critical significance of accurate rainfall forecasting in domains like agriculture, water resource management, and flood prediction, particularly amidst evolving climate dynamics, this research sheds light on the intricate nuances of rainfall prediction through scrutiny of distinct machine learning techniques. The results were revealed that for hourly rainfall data analysis at Pejabat, the LSTM model had the best accuracy, while for Kajang and Petaling, the Linear Regression model was best depending on the geographic and temporal conditions of the catching area. The Gradient Boost Regressor was excellent at predicting Kajang’s daily rainfall, and the ensemble technique was sometimes better.