A HYBRID FORECASTING MODEL FOR GOLD PRICE PREDICTION

Abstract

Gold price forecasting remains a challenging task due to the nonlinear and dynamic nature of financial markets. While technical indicators are widely used to support investment decisions, their predictive performance is often limited when applied individually. This study proposes a hybrid forecasting framework that integrates traditional technical indicators with ensemble machine learning algorithms to improve the prediction of XAU/USD price movements. Historical daily gold price data spanning from July 2023 to March 2026 were collected, and a set of technical features, including the Relative Strength Index (RSI), Moving Average Convergence Divergence (MACD), MACD Signal, Bollinger Bands, and calendar-based variables, were engineered. Standalone technical indicator models were first evaluated as baseline benchmarks, followed by the implementation of Random Forest, XGBoost, and Gradient Boosting models. A 5-fold time series cross-validation strategy was employed to ensure robust model evaluation while preserving the chronological order of the data. The results show that ensemble machine learning models generally outperformed traditional technical indicator approaches. Among the baseline models, Bollinger Bands achieved the highest classification accuracy of 57.53%, while the proposed hybrid model combining Random Forest and Gradient Boosting achieved the best overall performance with an accuracy of 60.14%. Feature importance and SHAP analyses identified Bollinger Band variables, particularly the lower band (BB_Low), as the most influential predictors, emphasizing the importance of volatility-related information in gold price forecasting. The proposed framework demonstrates that combining technical indicators with ensemble machine learning techniques can improve directional prediction accuracy and provide a practical decision-support tool for financial market participants.