Machine learning classification and feature perturbation analysis of a public gastric cancer dataset

Authors: Sun, Haixiang; Zhang, Yu; Li, Sisi; Wu, Xiaojie; Chen, Song

Source journal: Computational biology and chemistry (2026-11-11)

DOI: 10.1016/j.compbiolchem.2026.109569

Abstract

This study examines machine learning classification and model sensitivity using a public dataset described as gastric cancer-related. The source file contains clinical, demographic, diagnostic, and miRNA-related attributes and a binary cancer-status label whose clinical provenance could not be independently verified. Logistic regression, Random Forest, Gradient Boosting, XGBoost, LightGBM, K-Nearest Neighbors, voting, and stacking classifiers were compared through internal evaluation. Numerical feature perturbations examined changes in predicted positive-class probability, while paired-feature analyses explored departures from an additive model-response reference. Exploratory clustering and numerical search examples were used to examine the structure and sensitivity of model predictions. The reported perturbation effects were generally small and varied across observations and modification strengths. These analyses characterize computational behavior within the dataset; they do not establish biological causality, clinically validated patient subgroups, therapeutic interventions, or drug-repurposing candidates. Interpretation is limited by uncertainties concerning dataset provenance and annotations and the absence of independent validation. Further evaluation on independently curated data is required before biological or clinical applications can be considered.

Read the full text

Related topics

Related open access articles

Abstract and bibliographic data from PubMed (U.S. National Library of Medicine). Copyright of the abstract belongs to the original publisher or authors.