A new multi-tree Genetic Programming approach to feature construction in high-dimensional classification

Created by W.Langdon from gp-bibliography.bib Revision:1.8396

@Article{Chen:2025:knosys,

• author = "Ke Chen and Mingyang Dao and Ying Bi and Jing Liang and Zhenlong Wu and Peng Wang",
• title = "A new multi-tree Genetic Programming approach to feature construction in high-dimensional classification",
• journal = "Knowledge-Based Systems",
• year = "2025",
• volume = "319",
• pages = "113643",
• keywords = "genetic algorithms, genetic programming, Feature construction, Feature selection, High-dimensional classification",
• ISSN = "0950-7051",
• URL = "https://www.sciencedirect.com/science/article/pii/S0950705125006896",
• DOI = "doi:10.1016/j.knosys.2025.113643",
• abstract = "Classification, as a key task in machine learning, has been widely studied. However, with the rapid advancement of information technology, features in classification tasks increasingly exhibit low discriminability and strong redundancy. And the classification data has gradually changed from low-dimensional data to high-dimensional data. These characteristics not only significantly increase the complexity of the classification model training, but also lead to a decline in the generalisation ability of the classification model. How to reduce the number of original features and improve their differentiation becomes the key to improve the classification accuracy. Feature construction stands out as a crucial data processing technique for enhancing data quality. Genetic Programming (GP) has found extensive applications in feature construction due to its flexibility and strong interpretability. However, the existing GP-based feature construction methods suffer from the interference of redundant and irrelevant features in the face of high-dimensional data. By eliminating irrelevant and redundant features, reducing the search space of GP can aid in constructing more discriminative features. Motivated by this, we propose a GP algorithm that combines feature selection with feature construction for high-dimensional classification. During the evolution of GP, a subtree archive is introduced to store promising subtrees and these subtrees assist the generation of offspring. The experimental results show that the proposed method can achieve better classification performance than single-tree and multi-tree GP methods",

}

Genetic Programming entries for Ke Chen Mingyang Dao Ying Bi Jing Liang Zhenlong Wu Peng Wang

Citations