Comparative Analysis of ARIMAX, Support Vector Regression, and Linear Regression for Rice Price Prediction Using Weather-Based Time Series Features

Abstract

Rice prices are an important indicator of economic stability and food security in Indonesia. Rice price fluctuations are influenced by various factors, including weather conditions, making accurate predictive models necessary to capture data patterns effectively. This study aims to compare the performance of ARIMAX, Support Vector Regression (SVR), and Linear Regression (LR) methods in predicting medium and premium rice prices using a time series approach with rainfall as an external variable. The data used consist of daily rice price data from 2023–2025 obtained from the Department of Industry and Trade of Central Java Province and rainfall data from NASA POWER. The study applies preprocessing, feature lag, and time-based validation (time-based split), while also using Naïve Forecasting as a baseline model. Model evaluation was conducted using MAE, RMSE, and R² Score. The results show that Naïve Forecasting achieved the best performance for both commodities. Among the proposed models, SVR demonstrated the best performance for medium rice with an MAE of 26.03 Rupiah/kg, RMSE of 33.57 Rupiah/kg, and R² of 0.85, while Linear Regression showed relatively good performance for premium rice with an MAE of 42.54 Rupiah/kg, RMSE of 62.74 Rupiah/kg, and R² of 0.92. Meanwhile, ARIMAX did not achieve optimal performance. The Wilcoxon Signed-Rank Test results indicate that the performance differences among models are statistically significant (p-value < 0.001). This study contributes through the implementation of a comparative time series evaluation framework that integrates feature lag, external variables, time-based validation, and baseline forecasting for food commodity price prediction.