Comparison of Hybrid CNN-LSTM, LSTM, and CNN Models for Stock Price Prediction (Case Study: PT. Indofood Sukses Makmur Tbk)

Abstract

This study develops a hybrid deep learning model by combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to predict the stock price of INDF.JK using historical data from 2015 to 2025. Feature extraction is performed using Conv1D, followed by MaxPooling1D to reduce dimensions, and LSTM to capture time-dependent patterns. The model is evaluated using the R², RMSE, MAE, and MAPE metrics. The CNN-LSTM model demonstrates the best performance with an R² of 0.9759, RMSE of 87.77, MAE of 63.97, and MAPE of 1.02%. As a comparison, the single CNN model produced an R² of 0.9711, RMSE of 96.18, MAE of 71.16, and MAPE of 1.12%, while the single LSTM model obtained an R² of 0.9752, RMSE of 89.13, MAE of 66.99, and MAPE of 1.07%. These results confirm that the hybrid approach is superior in terms of stock price prediction accuracy compared to the use of a single model.