Improving Stock Price Prediction Accuracy Through Deep Learning Models and Hidden Markov Models: A Comparative Analysis

Abstract

Stock price prediction is a crucial research area, especially in the financial market, which plays a critical role in the global economy. This research compares the forecasting performance of several deep learning models on the BSE dataset. The models used are Long Short-Term Memory (LSTM), CNN-LSTM, Recurrent Neural Network (RNN), and a hybrid model of CNN-LSTM and Hidden Markov Models (HMM). Standard performance measures, such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Square Error (RMSE), and the coefficient of determination (R²), are used to compare the models. The experimental results showed that the hybrid LSTM-HMM model achieved the best predictive performance among the approaches tested. The RNN model performed well in forecasting, while the CNN-LSTM and RNN-HMM models showed relatively weak performance. The results indicate that deep learning can be combined with probabilistic models like Hidden Markov Models to enhance the prediction accuracy and reliability of the stock price. Future research may focus on incorporating advanced optimization algorithms with these hybrid models to further enhance forecasting performance.