Analisis Komparatif Algoritma Machine Learning untuk Prediksi Produksi Padi Kabupaten Bandung

Abstract

Rice production is an important indicator for supporting regional food security, including in Bandung Regency. This study aims to compare the performance of four machine learning algorithms, namely Linear Regression, Support Vector Regression (SVR), Random Forest, and XGBoost, in predicting rice production based on harvested area and productivity data from 2018 to 2025. The data were processed through missing value checking, chronological ordering by year, and feature engineering using production lag, rolling mean, and the interaction between harvested area and productivity. Model validation was conducted using Leave-One-Out Cross Validation (LOO-CV) because the dataset was limited to eight annual observations. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The results show that Linear Regression achieved the best performance, with an MAE of 75,277 tons, RMSE of 207,530 tons, MAPE of 0.24%, and R² of 0.9931. Random Forest ranked second with a MAPE of 4.45% and R² of 0.5508. XGBoost produced a MAPE of 5.08% and R² of 0.4810, while SVR showed the lowest performance with a MAPE of 7.36% and R² of -0.1583. The rice production forecast for 2026-2030 indicates a stable trend, ranging from approximately 31.28 to 35.37 million tons depending on the model used. This study provides an initial basis for developing data-driven rice production prediction systems at the regency level.