Hybrid Neuro-Fuzzy Deep Learning with Genetic Optimization for Explainable Stock Price Forecasting in Emerging Markets

Abstract

Precise stock price forecasting is vital for economic stability and capital allocation, yet it remains a tenacious challenge in emerging economies due to the inherent uncertainty and non-linearity of financial time series. Despite advances in deep learning, existing models often lack linguistic interpretability, fail to adapt to rapid market shifts, or exhibit look-ahead bias due to static validation splits. Moreover, empirical research focused on African financial systems, such as the Nigerian market, remains sparse, limiting the practical utility of conventional black-box architectures. This study proposes a Hybrid Neuro-Fuzzy and Deep Learning (HNFDL) framework that integrates fuzzy inference systems with Long Short-Term Memory (LSTM) networks and Genetic Algorithms (GA). The objective is to unify semantic reasoning with temporal learning to improve forecasting accuracy while maintaining high model transparency through explainable AI (XAI). Empirical validation using data from the Nigerian Exchange Group (NGX) (Dangote Cement, Zenith Bank, and the NSE All-Share Index) shows that the HNFDL model achieved a directional accuracy of 68.4% and a Mean Absolute Percentage Error (MAPE) as low as 4.36%. An ablation study confirmed that GA-driven optimization reduced the Root Mean Square Error (RMSE) by 8.4%, while the Diebold-Mariano test () statistically confirmed the model's superiority over standalone LSTM and fuzzy baselines. These results demonstrate that combining explainable fuzzy reasoning with adaptive deep neural architectures significantly enhances decision-making confidence. The framework provides a robust, statistically validated decision-support tool for investors and policy makers operating within volatile, information-asymmetric financial environments.