Stock Price Prediction Based on CNN-LSTM Model with Bayesian Optimization

Stock price fluctuations are highly complex and non-linear, creating substantial difficulties in making precise predictions within the financial industry. As financial market data grows and computational advancements continue, traditional linear models fall short of delivering accurate predictions. Recently, deep learning frameworks like LSTM networks and CNNs have become prominent in stock price forecasting because of their exceptional capabilities in processing time series and high-dimensional datasets. However, integrating these models efficiently and optimizing their hyperparameters remains a challenging task. This study examines the effectiveness of LSTM, CNN, and hybrid models in stock price prediction, introducing Bayesian Optimization for fine-tuning the hybrid model's hyperparameters. Results from multiple stock indices show that the Bayesian-optimized LSTM-CNN hybrid model delivers superior prediction accuracy and stability, across loss functions like MSE, RMSE, MAE, and Huber Loss.