<![CDATA[Stock price forecasting is a challenging task due to the complex, nonlinear, and dynamic nature of financial time series data. This study aims to develop a hybrid forecasting model by integrating Seasonal and Trend Decomposition using Loess (STL) with Long Short-Term Memory (LSTM) and to compare its performance with the Seasonal Autoregressive Integrated Moving Average (SARIMA) model as a linear benchmark. The empirical analysis is conducted using daily closing price data of PT Semen Indonesia (Persero) Tbk (SMGR) over the period from March 2018 to March 2026. The proposed approach applies STL decomposition to separate the time series into trend, seasonal, and residual components, enabling the LSTM model to capture nonlinear patterns more effectively. Forecasting performance is evaluated using the Mean Absolute Scaled Error (MASE) on an out-of-sample testing dataset. The results show that the hybrid STL–LSTM model achieves superior accuracy, with a MASE value of 0.4738, significantly outperforming the SARIMA model, which yields a MASE value of 2.7073. In contrast, the SARIMA model produces overly smooth forecasts and fails to capture short-term fluctuations and nonlinear dynamics present in the data. These findings indicate that the integration of STL decomposition and LSTM provides a more effective and flexible framework for modeling complex financial time series. The proposed model not only improves forecasting accuracy but also produces stable and reliable predictions, making it suitable for practical applications in stock price forecasting.]]>
Copyrights © 2026
