Compact and Interpretable Binary Classification with SLIM Geometric Semantic Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.9184

@Misc{Nardone:2026:SSRN,

• author = "Emanuele Nardone and Alessandra {Scotto di Freca} and Claudio {De Stefano} and Francesco Fontanella and Leonardo Vanneschi",
• title = "Compact and Interpretable Binary Classification with {SLIM} Geometric Semantic Genetic Programming",
• journal = "SSRN",
• year = "2026",
• keywords = "genetic algorithms, genetic programming",
• abstract = "Genetic Programming provides a natural basis for interpretable classification models, but its application to classification remains limited by search instability, model growth, and the trade-off between predictive accuracy and transparency. This paper presents the first extension of the Semantic Learning algorithm with Inflate and Deflate Mutations (SLIM) to binary classification, transferring the previously proposed geometric semantic classification framework to SLIM's compact semantic representation. Classification is performed via a logistic output transformation optimized for RMSE during training, and binary predictions are obtained by thresholding the model output. The proposed framework is evaluated on 18 real-world binary classification datasets through four complementary experimental analyses, including the study of the sigmoid scaling parameter, hyperparameter sensitivity, feature-selection behavior, and comparisons with GSGP and nine established classification methods. The results show that classification performance is largely insensitive to the sigmoid scaling parameter, supporting sigma= 1.0 and SLIM+ SIG2 as robust default choices. Hyperparameter analyses reveal complementary roles among the main parameters, with maximum tree depth serving as an implicit structural regularizer that influences both feature coverage and selection stability. Across the benchmark suite, SLIM outperforms standard GP, GSGP, and most of the considered machine learning classifiers, remaining competitive with Random Forest while evolving smaller symbolic models. ...",
• notes = "University of Eastern Finland",

}

Genetic Programming entries for Emanuele Nardone Alessandra Scotto di Freca Claudio De Stefano Francesco R Fontanella Leonardo Vanneschi

Citations