ANALISIS PERBANDINGAN MODEL GRU DAN LSTM UNTUK PREDIKSI HARGA SAHAM BANK RAKYAT INDONESIA Deep Learning, GRU (Gated Recurrent Unit), LSTM (Long Short-Term Memory), Stock Price Prediction

Abstract

This research implements and compares two deep learning architectures, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), for predicting the stock price of Bank Rakyat Indonesia (BRI) using historical data from February 2023 to October 2024. Through systematic hyperparameter tuning and comprehensive evaluation, the study finds that GRU consistently outperforms LSTM across all regression metrics, with a 10.7% improvement in R² and an 18.5% reduction in MAPE. The optimal GRU configuration (100 units, 100 epochs, batch size 32, learning rate 0.001) achieves an MSE of 6517.5 and MAPE of 1.3764%. Visual analysis confirms GRU's superior ability to capture stock price fluctuations and adapt more quickly to trend changes. The simpler architecture of GRU with fewer parameters proves more effective for handling the high-noise characteristics and varying volatility of stock price data. While both models face challenges in predicting extreme market events, GRU demonstrates better resilience and faster recovery after such occurrences. This research contributes to the understanding of recurrent neural network applications in financial time series forecasting and provides practical insights for developing more accurate stock price prediction systems.