Comparison of RNN TCN and GRU in Stock Price Prediction Accuracy Using Artificial Neural Networks

Stock price prediction is a challenging task in financial markets, offering valuable insights for investors and stakeholders. Deep learning techniques, particularly Artificial Neural Networks (ANNs), have shown promise in addressing this challenge. This study presents a comparative analysis of three deep learning architectures: Recurrent Neural Networks (RNNs), Temporal Convolutional Networks (TCNs), and Gated Recurrent Units (GRUs) for stock price forecasting, using historical data from PT. Astra Internasional Tbk, a leading Indonesian automotive company. We trained and tested each model with historical stock data, evaluating performance using metrics like Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-Squared (R²). The results show that TCN outperforms RNN and GRU in prediction accuracy, with an MAE of 77, MSE of 11639, RMSE of 107, MAPE of 1.26%, and an R² value of 0.96. These findings suggest TCNs are well-suited for stock price forecasting, as they capture temporal dependencies more effectively than recurrent models. This research advances predictive models in financial markets, providing insights into selecting deep learning techniques for accurate stock price forecasts.