Heuristic Search Method Improves Cancer Classification Using Gene Expression Data

A newly developed feature selection method combining a local search algorithm with a genetic algorithm has demonstrated significant performance improvements in classifying cancer from high-dimensional gene expression data. The hybrid technique, designated LSFS plus GA, was designed to address high-dimensional classification challenges in bioinformatics by optimizing the selection of relevant genomic features.

To evaluate the method, the investigators analyzed more than 30,000 patient records gathered from six open datasets.

The preprocessing pipeline applied normalization, outlier elimination, and logarithmic transformation techniques across the analyzed patient datasets.

The evaluation framework utilized contemporary machine learning algorithms to assess the effectiveness of the proposed feature selection system. Benchmark classifiers tested alongside the LSFS plus GA method included random forests, support vector machines, and neural networks.

In comparative testing, the LSFS plus GA method achieved an 8 to 15 percent increase in classification accuracy, a 10 to 20 percent improvement in sensitivity, and a 5 to 10 percent gain in specificity across datasets. The model also yielded a 12 to 18 percent reduction in data noise and a decrease in entropy reflecting a 6 to 9 percent increase in classification confidence, with statistical significance confirmed via cross-validation and analysis of variance at p < 0.05, while substantially reducing computational complexity.

According to the researchers’ conclusions, this developed algorithm enables a substantial reduction in computational complexity while simultaneously maintaining, or even improving, classification accuracy. The authors emphasized that achieving lower computational complexity alongside sustained accuracy is particularly important when analyzing cancer data, as this balance helps ensure more reliable analytical results. Ultimately, the study demonstrated the practical potential of pairing local search techniques with genetic algorithms to resolve the distinct challenges associated with high-dimensional gene expression data in oncology.

Subject of Research: Heuristic feature selection for cancer classification using high-dimensional gene expression data

Article Title: New heuristic method for cancer classification using gene expression data

Article References: Gu, S., Savina, T., Monov, D., & Gorozhanina, E. (2026). New heuristic method for cancer classification using gene expression data. Cancer Imaging, 26(1), Article 151. https://doi.org/10.1186/s40644-026-01111-y

Image Credits: AI Generated

DOI: 10.1186/s40644-026-01111-y

Keywords: cancer classification, gene expression data, feature selection, local search algorithm, genetic algorithm, high-dimensional data, bioinformatics, random forests, support vector machines, neural networks, GEO database, TCGA database, blood cancer, breast cancer, colon cancer, computational complexity, classification accuracy, entropy reduction