Genetic programming-based automated feature construction for robust high-dimensional tabular data classification

Toni Arifin, Ignatius Wiseto Prasetyo Agung, Erfian Junianto, Metha Fatkhaturohmah

Abstract


High-dimensional tabular datasets, particularly messenger RNA (mRNA) expression profiles, pose challenges due to limited samples, redundant features, and complex non-linear relationships. Conventional feature selection methods, such as mutual information (MI), select existing features but cannot construct new representations to capture complex dependencies. This study proposes genetic programming-based automated feature construction (AFCGP) integrated with MI and support vector machine (SVM) (AFCGP+MI+SVM) for high-dimensional gene expression classification. GP generates symbolic features by combining original features through mathematical functions, while MI selects relevant constructed features before classification using an SVM with a radial basis function (RBF) kernel. Evaluation employed repeated stratified 5-fold cross-validation with 10 repetitions and the Wilcoxon signed-rank test. Using the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) dataset, AFCGP+MI+SVM achieved 99.94% accuracy, an F1-score of 1.00, and a mean balanced accuracy of 1.00 (SD=0.0026). Combining 10 constructed features with original features yielded high performance, whereas constructed features alone could not fully replace the original information. Statistical analysis revealed no significant performance differences among the evaluated approaches (p>0.05). This study contributes an integrated framework combining GP-based symbolic feature construction and post-construction MI selection, with explicit evaluation of feature representation quality and classification stability.

Keywords


Automated feature construction; Genetic programming; High-dimensional data; Mutual information; Support vector machine

Full Text:

PDF


DOI: https://doi.org/10.11591/eei.v15i6.13526

Refbacks

  • There are currently no refbacks.


Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Bulletin of EEI Statistics

Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191 , e-ISSN: 2302-9285
This journal is published by the Institute of Advanced Engineering and Science (IAES) .