Grouping and optimal decision timing based on XGBoost-dual-objective optimization under multi-feature fusion

Abstract

Optimizing decision timing is a core component of precise decision-making across multiple domains. However, traditional methods suffer from limitations such as over-reliance on a few core features, imbalance between cost and performance objectives, and weak resistance to interference. To address these issues, this paper constructs a hierarchical optimization framework based on “dual-feature synergy and multi-feature fusion,” proposing a decision timing optimization method that integrates XGBoost with dual-objective optimization. First, under the dual-core-feature scenario, the XGBoost binary classification model achieved a test set accuracy of 0.8738 and a recall rate of 0.99 for compliant samples. Dual-objective optimization yielded six optimal grouping sets and corresponding decision timepoints, revealing the pattern: “Higher Feature 1 values correspond to moderately delayed decision timepoints.” Second, in the multi-feature fusion scenario, five additional correlated features were incorporated to update the model, effectively reducing abnormal interference. Reusing the optimization framework yielded six optimal groupings with a composite fitness of 0.8811, revealing the pattern: “Early decision for medium-to-high Feature 1 values, moderate delay for low and extremely high Feature 1 values.” Finally, multi-gradient detection error tests ranging from 5% to 30% demonstrated that the core metrics of the multi-feature model degraded by less than 5%, with recall rates remaining stable above 0.988. Its interference resistance significantly outperformed the dual-feature model. This method achieves dual improvements in decision timing accuracy and robustness, providing quantitative decision references for different Feature 1 groups. It holds significant practical value for similar multi-feature decision optimization problems.