A Feature Ensemble Framework for Stock Market Forecasting Using Technical Analysis and Aquila Optimizer

Stock market prediction relies heavily on combining different features due to the complex factors affecting stock prices and varying datasets. This study introduces a new method for feature fusion that improves predictions for traders and investors. We focus on three key types of technical analysis: momentum, trend, and volatility, and combine them using four different fusion strategies. These strategies include combinative fusion-based feature set (CFFS), adaptive feature-weighted fusion-based feature set (AWFS), feature-type fusion-based optimized feature set (FTFOFS), and feature-based optimized fusion feature set (FOFFS). The Aquila optimization technique is used to enhance these feature sets, adjusting feature weights to improve accuracy. We tested the performance of these optimized feature sets using forecasting models like decision tree (DT), naive bayes (NB), support vector regression (SVR), and multi-layer perceptron (MLP). The effectiveness of our approach is compared with other optimization methods, such as genetic algorithm (GA) and particle swarm optimization (PSO), over a 10-year period (2012-2022) with data from State Bank of India (SBI) and ICICI Bank Ltd (ICBK). The models predict short-term stock movements (3, 7, and 15 days ahead), and we evaluate their performance using various metrics like mean absolute error (MAE) and correlation coefficient (R-2). Our results show that the FOFFS-Aquila method significantly improves the MLP's predictions compared to other models. We also provide insights into the efficiency of the MLP based on FOFFS-Aquila, including its statistical validity and execution time.