<![CDATA[Stock price forecasting is a crucial aspect of financial market analysis, particularly in supporting more accurate and informed investment decision-making. This study compares the performance of the statistical Autoregressive Integrated Moving Average (ARIMA) model with three variants of the Long Short-Term Memory (LSTM) architecture, namely Vanilla LSTM, Bidirectional LSTM, and Stacked LSTM, in predicting closing prices and trading volumes of Indonesian bank stocks—specifically BBCA.JK, BBRI.JK, and BMRI.JK. The data were obtained from Kaggle and processed through normalization, transformation, and model training stages using Google Colab and TensorFlow. Evaluation was conducted using RMSE, MAE, and MAPE metrics. The results indicate that ARIMA performs better in forecasting closing prices, achieving an average MAPE of 1.9%, while Bidirectional LSTM yielded the best results in forecasting trading volumes, particularly for BBRI and BMRI stocks. However, the prediction error for volume data remains relatively high (average MAPE of 36.4%) due to its volatile nature. These findings suggest that data characteristics significantly influence model effectiveness. LSTM-based models demonstrate superior capabilities in capturing complex non-linear patterns and exhibit advantages in multivariate forecasting compared to the ARIMA model. This study is expected to serve as a reference for selecting appropriate forecasting models in the context of Indonesian banking stock markets. The results highlight a trade-off between ARIMA, which excels in modeling linear patterns such as closing prices, and LSTM, which is more adaptive to non-linear patterns like trading volumes.]]>
Copyrights © 2025
