AI-Powered GRU untuk Prediksi Saham Syariah Indonesia Berbasis Analisis Multi Faktor
TL;DR
The results show that the GRU model achieves high predictive accuracy, with a MAPE of 0.0286 and an RMSE of 74.92, and visualizations including training vs validation loss, scatter plots, residual plots, and error distribution confirm that the model avoids overfitting and generalizes well.
Abstract
This study aims to develop a predictive model for Indonesian Sharia stock prices using the Gated Recurrent Unit (GRU) algorithm with a multi-factor analysis approach. The main challenge in analyzing Sharia stocks lies in the high volatility influenced by fundamental, technical, bandarmology, and macroeconomic factors. GRU was chosen because it has a simpler structure compared to LSTM while remaining effective in processing complex time-series data. The dataset includes variables such as EPS, PER, PBV, ROA, ROE, MA, RSI, MACD, foreign buy/sell, and the IHSG index, normalized using the Min-Max Scaler. The results show that the GRU model achieves high predictive accuracy, with a MAPE of 0.0286 and an RMSE of 74.92. Visualizations including training vs validation loss, scatter plots, residual plots, and error distribution confirm that the model avoids overfitting and generalizes well. Furthermore, the deployment of an interactive interface based on Gradio enables real-time prediction simulations, making this research not only academically significant but also practically useful for investors and policymakers in the Sharia capital market.