Prediction of the Stock Market Using a Hybrid Model Based on Feature Expansion and LSTM-Based Algorithms

Abstract

The complexity associated with predicting stock market trends has garnered significant attention from researchers and financial analysts. The widespread use of the stock market as an investment platform by individuals accentuates the importance of developing effective prediction strategies. This paper introduces a hybrid model (FRPS-Tech-BiBay) that integrates Fibonacci retracement values, Resistance, Pivot, and Support levels with Technical indicators and a deep learning model based on Bidirectional Long Short-Term Memory (BiLSTM), optimized through Bayesian optimization. This approach combines feature expansion with technical indicators and deep learning algorithms to enhance prediction performance. Features like differences, averages, technical values, trends, and patterns are created using mathematical formulas, technical indicators, and candlestick pattern identifiers. The expanded feature set is then modeled using the BiLSTM network and optimized through Bayesian methods. The model is evaluated using metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R2), demonstrating improved accuracy and reduced errors. By providing an expanded feature set to the BiLSTM model, we aim to improve both the efficiency and accuracy of the predictions for the next day’s closing price. The implementation of this strategy demonstrates a marked improvement in forecasting market prices compared to basic stock features.