Mutual Information-Based Evolutionary Feature Construction via Minimizing Redundancy and Maximizing Relevance

Created by W.Langdon from gp-bibliography.bib Revision:1.8528

@InProceedings{leng:2025:GECCOcomp,

• author = "Yunze Leng and Kei Sen Fong and Mehul Motani",
• title = "Mutual Information-Based Evolutionary Feature Construction via Minimizing Redundancy and Maximizing Relevance",
• booktitle = "Proceedings of the 2025 Genetic and Evolutionary Computation Conference Companion",
• year = "2025",
• editor = "Aniko Ekart and Nelishia Pillay",
• pages = "627--630",
• address = "Malaga, Spain",
• series = "GECCO '25 Companion",
• month = "14-18 " # jul,
• organisation = "SIGEVO",
• publisher = "Association for Computing Machinery",
• publisher_address = "New York, NY, USA",
• keywords = "genetic algorithms, genetic programming, evolutionary feature construction, mutual information: Poster",
• isbn13 = "979-8-4007-1464-1",
• URL = "https://doi.org/10.1145/3712255.3726727",
• DOI = "doi:10.1145/3712255.3726727",
• size = "4 pages",
• abstract = "In high-dimensional real-world datasets, a critical challenge for feature engineering (FE) in machine learning (ML) is that it must balance predictive accuracy with feature interpretability. This paper introduces a novel nested genetic programming (GP) approach that includes evolutionary feature construction integrated with mutual information (MI)-based feature selection. Our approach adopts a custom GP fitness function based on minimal-redundancy-maximal-relevance (mRMR) criterion, guiding the evolutionary process to construct features that maximize mutual information with target variables while minimizing redundancy among the constructed features. The proposed approach addresses the critical challenge mentioned above by improving predictive performance while yielding interpretable features that are backed by information theory principles. Extensive experiments across diverse datasets demonstrate that our approach, even when using a substantially reduced feature count, consistently outperforms both baseline methods using all original features and other contemporary FE techniques when compared at the same feature dimensionality. The effectiveness of our approach is validated across multiple ML models in regression tasks, showing improved accuracy and better generalization across diverse datasets and varying feature dimensionality.",
• notes = "GECCO-2025 GP A Recombination of the 34th International Conference on Genetic Algorithms (ICGA) and the 30th Annual Genetic Programming Conference (GP)",

}

Genetic Programming entries for Yunze Leng Kei Sen Fong Mehul Motani

Citations