Predicting Stock Price Movements in Volatile Markets: A Multi-Model Fusion Approach

The dynamic financial markets where forecasting stock prices poses a persistent challenge, many recent researches in finance have focused on deep learning-based stock price prediction. Our model proposes an innovative Multi-Model Fusion Deep Learning approach that intricately integrates stock and news data to enhance predictive accuracy. Recognizing the intricate interplay of various variables in stock market dynamics, the method underscores the importance of considering market volatility for precise predictions. Capitalizing on the rapid assimilation of information into stock prices, a ground breaking fusion technique is introduced, seamlessly combining features extracted from both stock price data using a customized architecture inspired by LSTM and textual news data through a specialized NLP. The key to this approach lies in a sophisticated fusion layer, where distinct yet complementary features converge, enabling the model to synthesize a comprehensive understanding of market behavior. This hybrid architecture, finely tuned through joint training, demonstrates its effectiveness through comprehensive experimentation, assessing its performance in predicting stock price movements. The findings of our model are its higher accuracy which is 93.84%, 90.9% precision, 93.02% F1 score, 95.23% Recall and 0.006 Mean squared Error (MSE) which is lower in prediction of stock movements. By seamlessly blending insights from diverse data sources, this innovative approach charts a promising path toward refining stock price predictions in the volatile landscape of financial markets, offering potential improvements in forecasting accuracy and decision making.