A Hybrid CNN-LSTM Approach for Effective Stock Price Prediction in Optimizing Investment Strategies

Stock price prediction is important for crafting optimal investment strategies in the financial sector. Traditional models like Autoregressive Integrated Moving Average (ARIMA), often used for their predictive simplicity, struggle with the dynamic nature of stock markets due to their linear constraints. This study explores a hybrid approach combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, utilizing their strengths to enhance prediction accuracy in various market conditions. For this analysis, twelve stocks representing diverse market performances over the past year were selected. These stocks were trained on nine years of historical data—incorporating daily open, low, high prices, and percentage changes as key features. The training involved 800 cycles per stock, each running 500 epochs, with varying hyperparameter combinations to optimize the model. The evaluation focused on the minimum mean absolute error recorded as the test loss and the mean absolute percentage error to assess precision. Results revealed that the CNN-LSTM model generally predicts stock prices effectively, with a minimum mean absolute percentage error of 0.00791. However, challenges arose with certain stocks, particularly those subject to abrupt price surges and external influences, which were less predictable despite hyperparameter adjustments. This analysis not only highlights the model’s abilities but also underscores the influence of external factors on prediction accuracy. The study of hyperparameters further demonstrated that while most stock prices are predictable, some remain challenging due to these unpredictable elements.