Prediksi Harga Pangan di Kota Pangkalpinang Menggunakan Algoritma XGBoost

Abstract

Food price fluctuations are a strategic issue affecting food security and regional inflation stability. Pangkalpinang City experiences significant inflationary pressure in the food sector, highlighting the need for accurate predictive models. This study aims to develop a time series–based price prediction model for strategic food commodities using the Extreme Gradient Boosting (XGBoost) algorithm. The dataset consists of daily prices of 40 food commodities obtained from the Pangkalpinang City Trade Information System, covering the period from January 1, 2024, to May 28, 2025. The research methodology includes data preprocessing, feature engineering, time series–based data splitting, hyperparameter optimization using Optuna, and model evaluation. Feature engineering incorporates price lag features, statistical features, calendar features, and national holiday indicators. Model performance is evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and the R-Squared (R²). The results indicate that XGBoost achieves high prediction accuracy for most commodities, particularly those with stable price patterns, while still capturing fluctuation trends in highly volatile commodities. The findings suggest that commodity-specific modeling provides better performance than a single global model and can support regional inflation control policies.