An interpretable neuro-fuzzy approach to stock price forecasting

Stock price prediction is a complex and difficult task due to the chaotic behavior and high uncertainty in stock market prices. The design of a highly accurate, simple and intelligible forecasting model is of prime importance in this field. With this aim, a number of research studies have employed fuzzy rule-based systems for stock price forecasting. But the main focus has been on obtaining fuzzy systems with high accuracy and the interpretability aspect has been overlooked due to the assumption that the fuzzy rule-based systems are implicitly interpretable in the form of fuzzy rules which is not essentially true. This paper proposes an efficient and interpretable neuro-fuzzy system for stock price prediction using multiple technical indicators with focus on interpretability–accuracy trade-off. The interpretability of the system is ensured by: (1) rule base reduction via selection of the best rules using rule performance criteria to obtain an efficient and a compact rule base which is easily comprehendible and (2) constrained learning during model optimization stage so that simple constraints are imposed on the updates of fuzzy set parameters due to which the system remains interpretable and forecasting accuracy is not compromised. For experimental evaluation of the proposed system, daily stock data of Bombay Stock Exchange, CNX Nifty and S&P 500 stock indices are used. The simulation results show that the proposed system obtains a better balance between accuracy and interpretability than two other artificial intelligence techniques and two statistical techniques commonly used in stock price prediction.