Market Regime Classification and Entry Window Prediction for IDX Blue Chips Using Random Forest and BiLSTM Models

This study evaluates market regime classification and entry window prediction for five IDX blue-chip stocks using Random Forest and Bidirectional LSTM models. Regimes are labeled from volatility and trend features derived from daily prices since 2004. Using an 80% time-based split, BiLSTM outperforms Random Forest in regime classification (ROC-AUC 0.919 vs. 0.729). Entry window prediction is harder, with BiLSTM achieving 0.743 ROC-AUC and limited gains from ensembling. Results show temporal modeling improves regime detection, while entry signals remain sensitive to label design and class imbalance.