An adaptive and enhanced framework for daily stock market prediction using feature selection and ensemble learning algorithms

Even a slight increase in accuracy when predicting the direction of stock movements can have a significant impact on the rate of returns. However, determining the most suitable variables, methods, and parameters to predict price changes is extremely challenging due to the multitude of variables influencing these changes. This paper presents an innovative prediction framework that combines ensemble learning and feature selection algorithms to effectively capture daily stock movements. The study focuses on predicting the change between the opening and closing prices of the subsequent day and employs a daily sliding window cross-validation methodology. The framework comprises fourteen variable groups encompassing a range of financial and operational indicators. Experimental findings indicate that a competitive performance was achieved for stocks within the Borsa Istanbul 30 index. Light Gradient Boosting Machines and Shapley Additive Explanations emerges as the optimal model combination and exhibits superior performance compared to a buy-and-hold strategy.