Stock Market Prediction Using a Hybrid Neuro-fuzzy System

Abstract

Stock market prediction is an important area of financial forecasting, which is of great interest to stock investors, stock traders and applied researchers. Main issues in developing a fully automated stock market prediction system are: feature extraction from the stock market data, feature selection for highest prediction accuracy, the dimensionality reduction of the selected feature set and the accuracy and robustness of the prediction system. In this paper, an automated decision tree-adaptive neuro-fuzzy hybrid automated stock market prediction system is proposed. The proposed system uses technical analysis (traditionally used by stock traders) for feature extraction and decision tree for feature selection. Dimensionality reduction is carried out using fifteen different dimensionality reduction techniques. The dimensionality reduction technique producing the best prediction accuracy is selected to produce the reduced dataset. The reduced dataset is then applied to the adaptive neuro-fuzzy system for the next-day stock market prediction. The neuro-fuzzy system forms the stock market model adaptively, based on the features present in the reduced dataset. The proposed system is tested on the Bombay Stock Exchange sensitive index (BSE-SENSEX). The results show that the proposed hybrid system produces much higher accuracy when compared to stand-alone decision tree based system and ANFIS based system without feature selection and dimensionality reduction.