Enhancing Stock Market Predictions Using Hybrid Deep Learning Models with Sentiment Analysis and Feature Engineering

Abstract

Stock price prediction remains a complex yet crucial task in financial markets, requiring robust methodologies to capture intricate dependencies in stock price movements. This study proposes the Hybrid Deep Learning for Stock Market Prediction (HDL-SMP) model, integrating Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory (BiLSTM), and an Attention Mechanism to enhance predictive accuracy. The model processes three key data sources: historical stock prices, technical indicators, and sentiment analysis scores from financial news and social media. The CNN layer extracts spatial patterns in price movements, while the BiLSTM layer learns long-term dependencies and captures both past and future trends. The Attention layer assigns dynamic weights to features based on their significance, optimizing prediction outcomes. Experimental evaluations on benchmark stock market datasets demonstrate the superiority of HDL-SMP over traditional machine learning models and existing deep learning architectures. The model achieves higher prediction accuracy, lower error rates, and improved robustness against market volatility. A comparative analysis with Support Vector Regression (SVR), CNN-BiLSTM, and other hybrid deep learning models confirms the effectiveness of HDL-SMP in financial forecasting. These findings highlight the potential of deep learning-based models in stock market prediction, aiding investors and financial analysts in making informed decisions.