Enhancing Weather Prediction Models through the Application of Random Forest Method and Chi-Square Feature Selection

Abstract

This study discovers weather forecast methodologies, concentrating mainly on the climatic issues faced by Indramayu Regency and its considerable impact on agriculture, specifically rice production and national food security. The study emphasizes the crucial need for accurate weather forecasting, especially in the context of ongoing climate change, by highlighting the region's vulnerability to weather anomalies and their possible disruption of crop output. To solve these issues, the study investigates machine learning techniques, particularly ensemble learning methods such as Random Forest in conjunction with Chi-Square feature selection. The article thoroughly outlines the research approach, including data collection from Indonesia's Meteorology, Climatology, and Geophysics Agency (BMKG), data pre-processing, feature selection processes, and data splitting. Notably, the methodology integrates the Synthetic Minority Over-sampling Technique (SMOTE) to adjust imbalanced data and uses key weather attributes for model construction (humidity, wind speed, and direction). The resulting Random Forest model performs well, with an accuracy rate of 87.6% in forecasting different types of rainfall. However, the study indicates potential overfitting in some rainfall classes, implying the need for additional data augmentation or modeling technique refining. In conclusion, this study demonstrates the potential efficacy of ensemble learning techniques in weather prediction, focusing on the Indramayu Regency. It emphasizes the need for exact forecasts in the agricultural and fisheries industries and suggests possibilities for additional investigation, such as research into alternative prediction approaches such as deep learning.