Gregorius Daniel Jodan Perminas
Jurusan Informatika, Universitas Kristen Duta Wacana

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Perbandingan Kinerja Random Forest dan Gradient Boosting pada Strategi Trading Saham Sektor Energi Berkapitalisasi Besar di Indonesia Gregorius Daniel Jodan Perminas; Michael Chandra
Jurnal Terapan Teknologi Informasi Vol 10 No 2 (2026): Jurnal Terapan Teknologi Informasi
Publisher : Fakultas Teknologi Informasi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.21460/jutei.2026.102.483

Abstract

This study compares the performance of Random Forest and Gradient Boosting in developing machine learning-based trading strategies for large-cap energy sector stocks in Indonesia. The analyzed stocks consist of BYAN, DSSA, CUAN, BUMI, ADRO, ADMR, AADI, TCPI, PTRO, and PGAS, selected from the IDNFinancials energy sector list in May 2026. Daily historical data were obtained from Yahoo Finance for the period from 2020 to April 2026. Technical indicators were constructed from momentum, volatility, and trend groups, while feature selection was performed exclusively on the training data using Spearman rank correlation. The final features selected for all stocks were RSI_14, BB_LOWER, and EMA_20. The data were divided chronologically into training, evaluation, and backtesting periods with an approximate 70:20:10 proportion and a five-trading-day purging gap between consecutive periods. Hyperparameter tuning was conducted using five-fold TimeSeriesSplit with a five-observation gap. Random Forest and Gradient Boosting showed relatively similar classification performance, with mean accuracies of approximately 0.528 and 0.531, respectively, indicating that neither model consistently dominated across all stocks. The best financial performance was obtained by Gradient Boosting on TCPI, producing a return of 61.578%, a win rate of 72.727%, and 11 trades, although the result remained below the Buy and Hold return of 103.540%. Sensitivity analysis further showed that classification performance and trading performance did not necessarily move in the same direction. These findings indicate that classification accuracy alone is insufficient for evaluating machine learning-based trading strategies and should be complemented by financial performance and robustness analysis.