Deep Ensemble Model for Stock Price Forecasting Using Technical Indicators and Bayesian Optimization
Year: 2026
Type: conference-paper
Abstract: Predicting stock prices in emerging markets like Indonesia is difficult because of their high volatility and complex, nonlinear behavior. As a solution, the given work proposes an ensemble of deep learning models with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), trained using a Bayesian meta-learning model, to enhance accuracy and interpretability. This model employs fifteen important technical indicators based on eighteen original features that were narrowed down to fifteen with the help of correlation analysis, testing the Variance Inflation Factor (VIF), and permutation-based feature importance. Trading data of two prominent Indonesian banking stocks, which include BBRI.JK and BBCA.JK, spanning fifteen years daily, was trained and evaluated. The ensemble performed well and, according to the results of the Bayesian hyperparameter tuning, the Mean Absolute Percentage Error (MAPE) values of 1.21% and 1.45% with BBCA and BBRI, respectively, were lower than any other baseline models. These findings indicate that the suggested interpretable ensemble is a robust, data-driven method of predicting stock prices in volatile markets and can be useful for finance-related decision-making and market analysis.
Authors: Rifki Hidayat, Suharjito
Institution: Binus University
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7201878403