Article Details

Abstract

Feature selection aims to identify the most representative subset of features from high-dimensional raw data to enhance model predictive performance, reduce computational complexity, and effectively mitigate overfitting. However, traditional feature selection methods exhibit significant limitations in capturing complex nonlinear interaction relationships. This paper proposes an integrated feature selection method, namely IFGC-FS, which combines interaction screening with graph clique search. The method first performs preliminary screening of main-effect features based on Kendall's correlation coefficient, then constructs a weighted feature graph using the SHAP interaction matrix, and finally extracts feature subsets with strong interaction relationships through a maximum clique search strategy. Experimental results on six publicly available benchmark datasets demonstrate that IFGC-FS achieves lower root mean squared error than comparative methods in both random forest and partial least squares regression models. Further Friedman test results indicate that IFGC-FS can significantly reduce feature dimensionality while maintaining or even improving predictive performance, thereby validating its effectiveness in uncovering complex feature interaction structures and enhancing model interpretability and generalization ability.

Keywords

feature selection; regression modeling; Kendall correlation; SHAP interaction; maximum clique

References

1. [1] Li Zhiqin, Du Jianqiang, Nie Bin, etc Overview of Feature Selection Methods [J]. Computer Engineering and Applications, 2019, 55 (24): 10-19
2. [2] Peng H C，Long F H，Ding C. Feature Selection Based on Mutual Information Criteria of Max-dependency, Max-relevance, and min-redundancy [J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2005，27（8）：1226-1238.
3. [3] Battiti R. Using mutual Information for Selecting Features in Supervised Neural Net Learning [J].IEEE Transactions on Neural Networks，1994，5（4）：537-550.
4. [4] Zeng Z L，Zhang H J，Zhang R，et al. A novel Feature Selection Method Considering Feature Interaction [J].Pattern Recognition，2015，48（8）：2656-2666.
5. [5] Hu L，Gao W F，Zhao K，et al. Feature Selection Considering two Types of Feature Relevancy and Feature Interdependency [J].Expert Systems with Applications，2018，93：423-434.
6. [6] Scheetz T E，Kim K Y A，Swiderski R E，et al. Regulation of Gene Expression in the Mammalian Eye and its Relevance to Eye Disease [J].Proceedings of the National Academy of Sciences，2006，103（39）：14429-14434.
7. [7] Bühlmann P，Kalisch M，Meier L. High-dimensional Statistics with a View Toward Applications in Biology [J].Annual Review of Statistics and Its Application，2014，1（1）：255-278.
8. [8] Kendall M G. A New Measure of Rank Correlation [J].Biometrika，1938，30（1/2）：81-93.
9. [9] Fan J Q，Lyu J C. Sure Independence Screening for Ultrahigh Dimensional Feature Space [J].Journal of the Royal Statistical Society Series B：Statistical Methodology，2008，70（5）：849-911.
10. [10] Lundberg S，Lee S I. A Unified Approach to Interpreting Model Predictions [PP/OL].（2017-05-22）.https://arxiv.org/abs/1705.07874.
11. [11] Hagberg A，Swart P J，Schult D A. Exploring Network Structure, Dynamics, and Function Using NetworkX [R].Los Alamos National Laboratory(LANL)，2007.
12. [12] Zhong W X，Liu Y W，Zeng P. A Model-free Variable Screening Method Based on Leverage Score [J].Journal of the American Statistical Association，2023，118（541）：135-146.
13. [13] Pan W L，Wang X Q，Xiao W N，et al. A Generic Sure Independence Screening Procedure [J].Journal of the American Statistical Association，2019，114（526）：928-937.

Funding

National Natural Science Foundation of China (82260849); Development Plan for University level Science and Technology Innovation Team of Jiangxi University of Traditional Chinese Medicine (CXTD22015); Discipline Construction Fund of Jiangxi University of Traditional Chinese Medicine in 2024 (2024jzzdxkx026)

About the authors

Zheng Shuifei (1998-), male, Han ethnicity, from Shangrao, Jiangxi, currently pursuing a master's degree in data mining and traditional Chinese medicine informatics; Corresponding author: Nie Bin (1972-), male, Han ethnicity, from Xiajiang, Jiangxi, professor, CCF member, PhD, research direction: data mining, traditional Chinese medicine informatics; Zhang Zhiyao (2001-), female, Han ethnicity, from Nanchang, Jiangxi Province, is currently pursuing a master's degree. Her research interests include data mining and traditional Chinese medicine informatics; Li Xiang (2001-), male, Han ethnicity, from Tongling, Anhui Province, is currently pursuing a master's degree in data mining and traditional Chinese medicine informatics.