Gold Price Forecasting with a Hybrid Queuing Search Algorithm and Extreme Gradient Boosting Regressor Model

Abstract

Gold is a conventional haven metal in wealth management and risk diversification tactics since its price trends mostly display signs of future political and economic trends. The forecasting of future prices of gold in the process of regression analysis is extremely challenging, considering their unpredictability and volatility. A hybrid model for a machine learning model is formed in this study by combining the Extreme Gradient Boosting Regressor (XGBR) with optimization techniques Mayfly algorithm (MA) and Queuing Search Algorithm (QSA). The hybrid model is developed in this study by using daily data points from 2011 to 2023, based on the variables Exponential Moving Average (EMA), Simple Moving Average (SMA), On- Bal Volume (OBV), Average True Range (ATR), along with Open, High, Low, volume, and Close patterns. 5-fold Cross-Validation is applied to test the performance of the model on 80% of the data used for training and 20% data used for testing. The QSA-XGBR model gives a better R² statistic with a value of 0.998 on the training data and 0.997 on the test data compared to other models, including Decision Tree (DT), Transformer, iTransformer, LSTM, XGBR, MA-XGBR, and Bi-LSTM with regard to accuracy. The SHAP value of each variable explains the importance of the variable. The variables with the highest contribution to the model are the Low variable, the High variable, and the Open variable. Additionally, this research assesses the generalization of the QSA-XGBR model by applying it to silver and crude oil, demonstrating its robust performance across multiple asset classes. Finally, this study demonstrated the significance of the improvement in forecast accuracy gained through the proposed model through the Diebold-Mariano test. For analysts and investors, the QSA-XGBR model provides insightful information that could be improved by macroeconomic variables and testing under volatile circumstances.