Predictive Model Construction Based on SSA-XGBoost Algorithm

In this paper, we propose an intelligent prediction model based on SSA-XGBoost, focusing on the synergistic optimization mechanism of Sparrow Search Algorithm (SSA) and Extreme Gradient Boosting Regression Tree (XGBoost). First, the input data are preprocessed in a standardized way to eliminate the feature scale differences, and the 70%-30% ratio is used to divide the training set and test set. Second, the initial model of XGBoost is constructed, the loss function is optimized by second-order Taylor expansion, and the regularization term is introduced to control the model complexity. Further, the SSA algorithm is used to dynamically optimize the key hyperparameters, including the maximum tree depth, the minimum number of leaf node samples and the number of iterations, and the fitness function is used to guide the parameter search to improve the model generalization ability. The experimental results show that the optimized SSA-XGBoost model significantly outperforms the benchmark model in terms of MAE, RMSE and R², in which the MAE is reduced by 20.44% and the R² is improved to 0.9216, which verifies its superiority in nonlinear high-dimensional data prediction. The model provides an efficient solution for accurate prediction of complex systems by combining adaptive parameter optimization and integrated learning.