MACHINE LEARNING UNTUK PREDIKSI HARGA SAHAM DENGAN VALIDASI STRATEGI TRADING MENGGUNAKAN BACKTESTING PADA SEKTOR PERBANKAN DI BURSA EFEK INDONESIA

Nurkalam, Haikal
Sulaiman, Maulana Muhamad

CENDEKIA : Jurnal Ilmu Pengetahuan
Vol. 6 No. 2 (2026)

Stock investment in the Indonesian banking sector faces challenges in predicting volatile price movements, especially for novice investors with limited literacy and experience. This research aims to build a banking stock price prediction system using the Random Forest algorithm with trading strategy validation through automatic backtesting. The system was developed using 22 technical indicators (Moving Average, RSI, MACD, Bollinger Bands, and price action) to predict stock prices of Indonesia's five largest banks (BBCA, BBRI, BMRI, BBNI, BBTN) across multiple prediction horizons (1, 2, 3, 5, 10, and 30 days). Evaluation was conducted on 30 combinations (5 stocks × 6 horizons) using MAPE, MAE, RMSE, and R² score metrics. Results show the model achieves excellent accuracy with an average MAPE of 2.05% for short-term predictions (H+1) and 5.27% for medium-term predictions (H+5), categorized as "Excellent". The model-based trading strategy outperforms Buy & Hold with an average outperformance of +19.32%, where BBCA shows the best performance with ROI of +22.6% and Sharpe Ratio of 2.20. Feature importance analysis reveals the dominance of price action (41.56%) and moving averages (35.73%), consistent with banking stock characteristics that follow medium-term trend patterns. The web-based system with intuitive visualization and automatic backtesting proves effective as a decision support tool to help novice investors make more managed investment decisions and improve capital market literacy. ABSTRAKInvestasi saham sektor perbankan di Indonesia menghadapi tantangan dalam memprediksi pergerakan harga yang volatil, terutama bagi investor pemula dengan keterbatasan literasi dan pengalaman. Penelitian ini bertujuan membangun sistem prediksi harga saham perbankan menggunakan algoritma Random Forest dengan validasi strategi trading melalui backtesting otomatis. Sistem dikembangkan menggunakan 22 indikator teknikal (Moving Average, RSI, MACD, Bollinger Bands, dan price action) untuk memprediksi harga saham lima bank terbesar Indonesia (BBCA, BBRI, BMRI, BBNI, BBTN) pada berbagai horizon prediksi (1, 2, 3, 5, 10, dan 30 hari). Evaluasi dilakukan pada 30 kombinasi (5 saham × 6 horizons) menggunakan metrik MAPE, MAE, RMSE, dan R² score. Hasil penelitian menunjukkan model mencapai akurasi sangat baik dengan MAPE rata-rata 2.05% untuk prediksi jangka pendek (H+1) dan 5.27% untuk prediksi jangka menengah (H+5), tergolong kategori "Excellent". Strategi trading berbasis model mengungguli Buy & Hold dengan rata-rata outperformance +19.32%, dimana BBCA menunjukkan performa terbaik dengan ROI +22.6% dan Sharpe Ratio 2.20. Analisis feature importance mengungkap dominasi price action (41.56%) dan moving averages (35.73%), sesuai karakteristik saham perbankan yang bergerak mengikuti pola tren jangka menengah. Sistem berbasis web interface dengan visualisasi intuitif dan backtesting otomatis terbukti efektif sebagai decision support tool untuk membantu investor pemula mengambil keputusan investasi yang lebih terkelola dan meningkatkan literasi pasar modal.