Following the collapse of the Bretton Woods system, the pricing of gold shifted from official to market-based mechanisms. Due to the unique attributes and historical significance of spot and futures gold, alongside influences from political, economic, and other factors, gold prices have remained volatile, impacting various financial activities worldwide. Predicting gold prices using time series and machine learning models, as well as various combination models, has become a longstanding focus. However, Given the multifaceted and nonlinear nature of the gold market, while combination models may potentially offer higher accuracy, they often require high technical expertise and lack generalizability, leading to difficulties in modeling, replicating, and low efficiency and timeliness. In the rapidly changing financial markets, their practical value to investors and enterprises is limited. This paper aims to select the optimal basic model to provide investors and enterprises with a sincere foundation for forecasting and outline a framework for developing combination models. Based on daily closing prices of spot and futures gold from January 3, 2023, to March 5, 2024, from investing.com, the test set consists of data from the past fifteen days, while the remaining data comprises the training set. The findings indicate a projected continuation of rising gold prices. And among the ARIMA, LSTM, and SVM models, the SVM model demonstrates relatively high accuracy with lowest MAPE, RMSE and MAE for both gold spot and gold futures so SVM is suggest being prioritized for predictions and hybrid model construction.