Advances in Industrial Engineering

Advances in Industrial Engineering

Developing a bat-based Boosted Neural Network Ensemble model for Stock price time series prediction

Authors
1 Department of Industrial Engineering, Faculty of Computer and Industrial Engineering, Birjand University of Technology, Birjand, Iran
2 Associate Professor, Department of Industrial Engineering, Arak University, Arak, Iran
Abstract
Stock price forecasting remains challenging because of financial markets' nonlinear, dynamic, and volatile behavior. This study proposes the Bat-Based Neural Network Ensemble (BBNNE) for multi-step-ahead forecasting. BBNNE integrates three components: (i) Relevant Lag Space Generation (RLSG), a tournament-based method that uses time-series autocorrelation to construct diverse lag subspaces; (ii) Bat Algorithm optimization of neural-network experts on their respective subspaces; and (iii) APE-thresholded boosting, which adaptively reweights training observations using a self-tuning error threshold. By jointly manipulating feature and instance spaces, the framework balances expert diversity and predictive accuracy. BBNNE is evaluated on four equities from contrasting markets: SHPETRO and FOLAD from the Tehran Stock Exchange and IBM and MSFT from the US market. Comparisons involve a multilayer perceptron (MLP), radial basis function network (RBF), Bagging, and AdaBoost. Across all datasets, BBNNE achieves the highest coefficient of determination (R²) and lowest horizon-aware Theil's U, with U below one only for BBNNE on IBM. It records the lowest mean absolute percentage error (MAPE) on FOLAD and IBM and performs comparably to the best baseline on SHPETRO and MSFT, without statistically significant differences. Paired Wilcoxon tests show significant improvements over at least one baseline on three datasets and no significant deterioration relative to any baseline. Ablation results confirm contributions from both RLSG and Bat Algorithm training, especially in the emerging-market series.
Keywords
Subjects


Articles in Press, Accepted Manuscript
Available Online from 17 November 2026