Application Of Support Vector Regression Method In Predicting Rice Prices In Langkat Regency

Abstract

This study aims to predict rice prices in Langkat Regency using Support Vector Regression (SVR) with a Radial Basis Function (RBF) kernel. Historical price data from 2020–2024 covering six rice categories in time series format were utilized. The research stages include data preprocessing, time-based feature engineering, chronological data splitting with an 80:20 ratio, SVR model training, and performance evaluation using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results demonstrate that SVR with RBF kernel achieves Very Good prediction accuracy across all categories, with MAPE values ranging from 5.07% to 7.01%, an average MAPE of 5.66%, MAE of IDR 779, and RMSE of IDR 834. The best performance was achieved by Medium Quality II rice (MAPE 5.07%), while Low Quality I rice exhibited the highest prediction error due to greater price fluctuations. Weather factor analysis revealed that daily temperature has a positive and statistically significant correlation with prediction error (r = +0.416 to +0.710, p < 0.001), where high temperatures consistently increased prediction deviation. Furthermore, feature engineering incorporating time-based features, lag features, and rolling mean proved effective in capturing price change patterns. The model was implemented into a web-based prediction system presenting performance visualization, weather analysis, and interactive prediction results, serving as a decision support tool for price control and food commodity planning in Langkat Regency.