Analisis Perbandingan Kinerja Model GRU Dan LSTM Dalam Prediksi Harga Saham BBCA

Abstract

This study comparatively analyzes the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models in predicting the closing price of PT Bank Central Asia Tbk (BBCA) stock using historical data from January 1, 2019 to April 30, 2026. Both models were evaluated using four metrics: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R²). The dataset consisted of 1,780 daily closing price records obtained from Yahoo Finance, processed through Min-Max normalization and windowing with a window size of 60 days. The LSTM model used three stacked layers (128→64→32 units) with 128,417 parameters, while the GRU model used an equivalent architecture with 96,993 parameters. Training was conducted for a maximum of 100 epochs with EarlyStopping and ReduceLROnPlateau callbacks. Results showed that GRU outperformed LSTM across all evaluation metrics, with RMSE 206.96 vs 255.45, MAE 163.56 vs 201.57, MAPE 2.09% vs 2.57%, and R² 0.9325 vs 0.8972. GRU also demonstrated better computational efficiency with shorter training time (54.87 vs 78.29 seconds). The GRU model was applied to forecast BBCA stock prices for May–August 2026, projecting a gradual recovery from Rp 6,166.56 to Rp 6,727.84.