Enhancing Sharia Stock Price Forecasting using a Hybrid ARIMA-LSTM with Locally Weighted Scatterplot Smoothing Regression Approach

Abstract

Predicting Sharia stock prices is complex because it has high volatility and non-linear data patterns. To improve the accuracy of the forecast, the right technique is needed according to the existing data pattern. One of the techniques currently developing is integrating (hybrid) two forecasting models. This study proposes a hybrid autoregressive integrated moving average (ARIMA) and long short-term memory (LSTM) model with the locally weighted scatterplot smoothing (lowess) linear regression technique. This model is designed by creating a linear regression between the actual value and the predicted results of the ARIMA and LSTM models using the Lowess technique. The dataset used here is the closing stock prices of four Indonesian Islamic banking companies. The hybrid ARIMA-LSTM model with lowess linear regression significantly outperforms the individual ARIMA and LSTM models because it produces better performance metrics, namely mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), for training and testing datasets. The proposed hybrid model effectively reduces noise, and the model can capture complex patterns in the Sharia stock price dataset, and the prediction results are more accurate. The accuracy values for training data and data testing datasets were respectively 97.6% and 98.3% (BANK. JK), 98.3% and 98.2% (BRIS. JK), 99.4% and 99.5% (BTPN. JK), and 97.7% and 99.3% (PNBS. JK).