Perbandingan Metode Random Forest dan LASSO untuk Nowcasting PDB Industri Manufaktur Indonesia

Abstract

Sektor industri manufaktur merupakan salah satu kontributor utama perekonomian Indonesia, namun publikasi data Produk Domestik Bruto (PDB) sektor ini sering mengalami keterlambatan. Penelitian ini bertujuan membandingkan kinerja metode random forest dan least absolute shrinkage and selection operator (LASSO) dalam pemodelan nowcasting PDB industri manufaktur Indonesia menggunakan data triwulanan periode 2012–2024. Analisis dilakukan melalui prapemrosesan data, analisis komponen utama, optimasi hiperparameter, dan walk forward validation. Hasil penelitian menunjukkan bahwa enam komponen utama mampu menjelaskan lebih dari 94% keragaman data. LASSO memberikan kinerja prediksi yang lebih baik dengan RMSE 14973.34 dan MAPE 2.23%, dibandingkan random forest dengan RMSE 28411.59 dan MAPE 4.28%. Hasil ini menunjukkan bahwa LASSO lebih efektif untuk nowcasting PDB industri manufaktur Indonesia.

The manufacturing sector is one of the main contributors to Indonesia’s economy. However, the publication of Gross Domestic Product (GDP) data is often delayed. This study compares the performance of random forest and least absolute shrinkage and selection operator (LASSO) in nowcasting Indonesia’s manufacturing GDP using quarterly data from 2012–2024. The analysis included data preprocessing, principal component analysis, hyperparameter optimization, and walk-forward validation. The results showed that six principal components explained more than 94% of the total variation. LASSO achieved better predictive performance with an RMSE of 14973.34 and a MAPE of 2.23%, whereas random forest produced an RMSE of 28411.59 and a MAPE of 4.28%. These findings indicate that LASSO is more effective for nowcasting Indonesia’s manufacturing GDP.