Gold Price Prediction Using a Hybrid LSTM and Prophet Model Enhanced with News Sentiment Analysis

Yessica Yamin (STMIK Time. Jalan Merbabu No. 32 aa-bb, Pusat Ps., Medan Kota, Medan City, North Sumatra 20212, Indonesia.)
Robet (STMIK TIME)
Hendri (STMIK TIME)

Jurnal Teknologi dan Manajemen Informatika
Vol. 12 No. 1 (2026): Juni 2026

This study aims to improve the accuracy of gold price forecasting by combining statistical and deep learning methods with sentiment analysis. Three models were developed and compared: (1) a pure Long Short-Term Memory (LSTM) model, (2) a hybrid LSTM + Prophet model, and (3) a hybrid LSTM + Prophet + Sentiment model. The datasets consisted of daily gold prices and financial news sentiment from 2013 to 2023. Each model was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared (R2), and Mean Absolute Percentage Error (MAPE). The pure LSTM model achieved an R2 of 0.9375, while the hybrid LSTM + Prophet model improved performance to 0.9394 with lower error rates. The integration of sentiment data resulted in stable but not significantly higher accuracy. Overall, the hybrid LSTM + Prophet model produced the best results, confirming that combining statistical trend decomposition with deep learning effectively enhances forecasting stability and interpretability for financial time series data such as gold prices.