Prediksi IHSG Berbasis Web Menggunakan Metode Long Short-Term Memory (LSTM)

Abstract

The Indonesia Composite Stock Price Index (IHSG) is one of the primary indicators used to measure the performance and stability of the capital market in Indonesia. The dynamic and non-linear characteristics of IHSG data make stock market prediction a challenging task when using conventional statistical methods. This study aims to develop a web-based IHSG prediction system using the Long Short-Term Memory (LSTM) method to improve prediction accuracy and provide an interactive forecasting platform for users. Historical IHSG data were collected from Yahoo Finance API using OHLCV (Open, High, Low, Close, Volume) variables. The data preprocessing stage included data cleaning, normalization using Min-Max Scaling, and sequence generation with the sliding window technique. Hyperparameter tuning was conducted by testing several configurations of window size, hidden units, learning rate, and epoch values. The best model configuration was obtained using a window size of 60, hidden units of 128, a learning rate of 0.01, and 74 epochs, resulting in RMSE of 51.9821, MAE of 39.8241, and MAPE of 0.559 %. Unlike previous studies that mainly focused on offline model evaluation, this research integrates the LSTM model into an interactive web-based prediction system equipped with visualization, AI forecasting, statistical evaluation, and batch prediction simulation features. The results indicate that the LSTM model is capable of producing accurate IHSG predictions and can be effectively implemented in a real-time web-based forecasting system.