COMPARATIVE ANALYSIS OF STATISTICAL AND MACHINE LEARNING MODELS FOR GOLD PRICE PREDICTION

Abstract

Gold remains one of the most important safe-haven assets, yet its volatile dynamics make accurate forecasting a persistent challenge. This study evaluates and compares statistical models (ARIMA, ETS, and Linear Regression) with machine learning approaches (KNN and SVM) using daily gold price data from 2021 to 2025, followed by forecasts for 2026. Descriptive statistics revealed moderate volatility (σ = 501.12) and strong historical growth (85% return), underscoring gold's strategic role in financial markets. Empirical results demonstrate that Linear Regression (R² = 0.986, RMSE = 35.7) and ETS outperform more complex algorithms, while KNN and SVM underperformed, often misrepresenting volatility. The 2026 forecast projects a mean gold price of $4,659, implying a 58.6% return, though risks from macroeconomic shocks remain. These findings highlight that transparent and interpretable models can surpass advanced machine learning in volatile markets, offering critical insights for investors, policymakers, and researchers in predictive financial analytics.