交互作用特征选择研究综述OA

Feature selection plays a crucial role in enhancing model performance,reducing computational cost,and imp-roving interpretability.Conventional methods mainly assess the relevance between features and target variables,often overlooking their potential interactions,which may cause the omission of key feature combinations.Therefore,researchers have proposed interaction-aware feature selection approaches.This paper systematically summarizes and compares main-stream modeling strategies from five perspectives:interaction order,representation form,structural relation,dependency type,and modeling strategy.It further introduces an"ideal-property"evaluation framework for interaction metrics and provides a comprehensive comparison of representative indicators.Finally,the paper discusses existing challenges and out-line future research directions,offering methodological insights for subsequent studies.