ANALISIS PERBANDINGAN MODEL SARIMAX, RANDOM FOREST, DAN HYBRID SARIMAX–RF PADA PERAMALAN CURAH HUJAN BULANAN DI KOTA MAJENE TAHUN 2025

Abstract

This study compares the performance of three forecasting models: SARIMAX, Random Forest (RF), and a Hybrid SARIMAX–RF approach in predicting monthly rainfall in Majene City for the year 2025. The dataset consists of five years of monthly rainfall and relative humidity data (2020-2024), sourced from BMKG and national meteorological records. The SARIMAX model captures seasonal trends and includes humidity as an exogenous variable, while the RF model addresses nonlinear relationships using lag-based predictors. A Hybrid model is then developed by combining SARIMAX outputs with RF-learned residual corrections. The forecasting performance is evaluated using standard error metrics such as MAE, RMSE, and MAPE on 2024 test data before forecasting 2025. Results show that the Hybrid model significantly outperforms the individual models, achieving the lowest MAPE (23.31%) compared to SARIMAX (42.21%) and RF (41.23%), and successfully capturing both seasonal patterns and irregular variations. The study demonstrates the importance of model integration for improving rainfall prediction accuracy, especially in tropical coastal regions with complex weather dynamics. Future studies are encouraged to explore additional Hybrid architectures and incorporate other climatic variables to further enhance predictive performance.