Gold is one of the most important investment commodities in the global financial system, widely recognized for its role as a safe-haven asset and its ability to preserve value during periods of inflation, economic instability, and geopolitical uncertainty. Despite its relative stability compared to other financial instruments, gold prices exhibit significant volatility driven by various macroeconomic factors, including exchange rate movements, inflation dynamics, global monetary policy decisions, and market sentiment. As a result, accurate gold price prediction remains a critical challenge for investors, financial analysts, and policymakers. This study aims to conduct a comparative performance analysis of two machine learning algorithms, namely Long Short-Term Memory (LSTM) and Support Vector Regression (SVR), in predicting gold prices represented by the XAU/USD currency pair. The research utilizes daily historical gold price data from 2004 to 2025 obtained from the Kaggle platform. The dataset includes key financial attributes such as Open, High, Low, Close prices, and trading Volume. Data preprocessing steps involve data cleaning, chronological sorting, handling missing values through linear interpolation, feature selection, and normalization using the Min-Max scaling technique. The dataset is then divided sequentially into training and testing sets with an 80:20 ratio to preserve temporal dependencies. The LSTM model is designed to capture long-term temporal patterns using the closing price as a time series input, while the SVR model leverages multiple input features to model non-linear relationships through kernel-based regression. Model performance is evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The experimental results demonstrate that the LSTM model outperforms the SVR model across all evaluation metrics. The LSTM achieved an RMSE of 0.0082, an MAE of 0.0060, and an R² value of 0.9969, indicating a very high level of predictive accuracy and strong generalization capability. In contrast, the SVR model recorded an RMSE of 0.0289, an MAE of 0.0143, and an R² of 0.9611, reflecting lower precision, particularly during periods of high price volatility. These findings confirm that LSTM is more effective in capturing complex temporal dependencies and non-linear dynamics inherent in gold price time series data. Consequently, LSTM is recommended as a superior approach for long-term gold price forecasting, while SVR may serve as a complementary or baseline predictive model in financial time series analysis.