Predictive Model for the Price of Gold Using Machine Learning

Reynafarje S.P.; Parra W.E.; Alarcon J.S.D. · International Conference on Artificial Intelligence, Computer, Data Sciences, and Applications, ACDSA 2026 (2026) · DOI: 10.1109/acdsa67686.2026.11467951

Resumen

The prediction of gold prices represents a significant challenge for sectors such as mining and finance, especially in emerging economies like Peru. Traditional models show limitations in the face of high market volatility and the presence of multiple influential variables, as they fail to capture nonlinear relationships and respond effectively to abrupt changes in the economic environment. In response to this issue, this study proposes a predictive model based on the XGBoost algorithm, using macroeconomic variables such as the dollar index, interest rates, oil prices, the Standard & Poor's 500 (S&P 500) index, and others related to international markets. The methodology was structured in four phases: dataset selection, algorithm training, evaluation using regression metrics, and temporal validation. The model was trained with historical data and validated with recent records to simulate real-world conditions. The results obtained in Experiment 2 indicate high predictive capacity, with a Mean Absolute Error (MAE) of 18.90 USD, Root Mean Squared Error (RMSE) of 23.45 USD, and a Coefficient of Determination (R2) of 0.92. Compared to the manual approach evaluated in Experiment 1, the model achieved a 29.6% improvement in predictive accuracy, a 93% reduction in response time, and a 29% increase in user satisfaction, establishing itself as an effective tool to support strategic decision-making in environments sensitive to gold price fluctuations.

Instituciones: Universidad Peruana de Ciencias Aplicadas