Kombinasi Inisialisasi Hyperparameter dan Algoritma Adam untuk Optimasi Model Gated Recurrent Unit dalam Meramalkan Harga Penutupan Saham Pasca Isu Boikot

This research aims to enhance the accuracy of daily closing price forecasting for PT Unilever Indonesia Tbk shares by optimizing a Gated Recurrent Unit (GRU) model, a deep learning architecture. The study focuses on the impact of boycott issues on company performance and explores the effectiveness of hyperparameter tuning and the Adam optimization algorithm. Utilizing a five-year historical dataset, time series analysis was employed. The results demonstrate that the GRU model with a configuration of 50 epochs, batch size 8, hidden state 64, and default Adam parameters achieved the lowest Mean Absolute Percentage Error (MAPE) of 1.59% and Root Mean Squared Error (RMSE) of 71.51 among the 81 configurations tested. While higher Adam parameter settings also yielded satisfactory results, this specific configuration exhibited superior performance. The findings highlight the sensitivity of model accuracy to sharp price fluctuations, suggesting that shorter forecast horizons may be more appropriate. This research contributes significantly to the advancement of accurate and reliable stock price forecasting models.