The dynamic modulus of asphalt mixture is an important parameter in the design of asphalt pavement. Extracting material characteristics, dynamic modulus, and phase angle information from a large amount of asphalt concrete datasets using integrated methods is of great significance for optimizing the performance of asphalt pavement. The extreme gradient boost (XGBoost) model aggregated a series of decision tree models through weighted summation to construct a powerful prediction model, while optimizing the loss function to minimize prediction errors. In order to further improve the accuracy of dynamic modulus and phase angle prediction, heuristic algorithms were used to optimize the model. Initially, the basic model was initialized based on samples and the gradient of the loss function of the training data was calculated. Subsequently, XGBoost utilized gradient details to construct a decision tree model, optimized leaf node weights, and updated the model’s predictions through weighted summation. During this process, heuristic algorithms are used to optimize the optimal parameters of the entire XGBoost model. The experimental results show that the improved XGBoost model outperforms the original model in all performance evaluation indicators, improving the accuracy of predicting the dynamic modulus and phase angle of asphalt mixtures.