Gold has long been considered a safe investment asset, especially during times of financial, economic, and political instability. To help investors plan future strategies, this research aims to build machine learning models to predict gold prices in Vietnam using data from 2010 to 2024, which incorporates a variety of macroeconomic and geopolitical factors. We applied the traditional ARIMA(SARIMAX) model as well as deep learning models such as Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Convolutional Recurrent Neural Network (CRNN) to predict gold price in the vietnamese market. The results show that SARIMAX achieved worse predictive accuracy (R2 = 0.7677, RMSE = 0.2342, MAE = 0.1933) compared to deep learning models, with the LSTM model performing the best (R2 = 0.9848, RMSE = 0.1250, MAE = 0.663). We further improved the SARIMAX by combining it with deep learning models using the stacking method, incorporating two meta learning models: Linear Regression (linear meta-model) and XGBoost (nonlinear meta-model). Experimental results found that the hybrid model using XGBoost outperformed the one using linear meta model (Linear Regression). Specifically, the best performance was achieved by combining SARIMAX with LSTM and meta model XGBoost, yielding R2 = 0.9903, RMSE = 0.0996, and MAE = 0.0581. Thus, by significantly improving the results, this hybrid model outperformed both SARIMAX and deep learning models. In summary, the application of Machine Learning and Artificial Intelligence algorithms in our research will bring clear benefits to investors and managers in making more accurate and effective gold investment decisions in the future.