PERBANDINGAN RANDOM FOREST REGRESSION DAN SUPPORT VECTOR REGRESSION DALAM PREDIKSI HOTSPOT DI RIAU DENGAN PENDEKATAN FEATURE ENGINEERING

Abstract

Kebakaran hutan dan lahan di Provinsi Riau merupakan ancaman tahunan yang memerlukan sistem peringatan dini berbasis prediksi hotspot. Penelitian ini membandingkan kinerja Random Forest (RF) dan Support Vector Regression (SVR) dalam memprediksi jumlah hotspot menggunakan pendekatan feature engineering dengan metrik evaluasi MAE dan RMSE. Peubah yang digunakan meliputi hotspot, curah hujan, anomali curah hujan, hari tanpa hujan, ENSO, dan IOD. Didapatkan hasil penerapan feature engineering efektif dalam meningkatkan performa Random Forest dan Support Vector Regression dengan penurunan nilai MAE dan RMSE yang signifikan. Melalui perbandingan kinerja pada kedua model didapatkan kinerja Random Forest lebih unggul dibandingkan Support Vector Regression yang ditunjukkan oleh nilai MAE dan RMSE yang lebih kecil.

Forest and land fires in Riau Province are an annual threat that requires an early warning system based on hotspot prediction. This study compares the performance of Random Forest (RF) and Support Vector Regression (SVR) in predicting the number of hotspots using a feature engineering approach, evaluated with MAE and RMSE metrics. The variables used include hotspots, rainfall, rainfall anomalies, consecutive dry days, ENSO, and IOD. The results show that the application of feature engineering is effective in improving the performance of both Random Forest and Support Vector Regression, as indicated by a significant reduction in MAE and RMSE values. Based on the performance comparison, Random Forest outperforms Support Vector Regression, as evidenced by lower MAE and RMSE values.