TY - JOUR AU - Bukhori, Hilmi Aziz AU - Bukhori, Saiful AU - Anam, Syaiful AU - Yusuf, Feby Indriana AU - Sari, Meylita PY - 2025/09/02 Y2 - 2025/11/14 TI - GWO-Enhanced Hybrid Deep Learning with SHAP for Explainable TLKM.JK Stock Forecasting JF - Jurnal Teknik Informatika (Jutif) JA - J. Tek. Inform. (JUTIF) VL - 6 IS - 4 SE - Articles DO - 10.52436/1.jutif.2025.6.4.5205 UR - https://jutif.if.unsoed.ac.id/index.php/jurnal/article/view/5205 SP - 2566-2585 AB - <p>This study presents an innovative Grey Wolf Optimization (GWO)-enhanced hybrid deep learning model integrating Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Transformer, combined with SHAP for interpretable stock price forecasting of TLKM.JK from July 29, 2024, to July 29, 2025. Addressing non-linear market dynamics, the model evaluates seven experimental cases, with the GWO-optimized configuration (Case 2) achieving superior performance, with a Root Mean Squared Error (RMSE) of 75.23, Mean Absolute Error (MAE) of 58.14, and Directional Accuracy (DA) of 76.2%, surpassing the baseline by 17.4% in RMSE and 8.1% in DA. Notably, Case 2 excels during the April 2025 surge (11.8% increase, MAE 53, DA 82%) and the high-volume day of May 28, 2025 (531,309,500 shares, MAE 48), leveraging Volume (SHAP 0.45) and RSI (0.28) as key predictors. With a 4-hour convergence time on an NVIDIA RTX 3060 GPU, the model ensures computational efficiency and interpretability, making it a robust tool for traders. Despite limitations in single-stock focus and GPU dependency, this framework advances AI-driven financial forecasting by offering transparent, high-accuracy predictions, paving the way for multi-stock applications and real-time SHAP updates.</p> ER -