MACHINE LEARNING-BASED CLASSIFICATION OF BREAST CANCER USING GENEEXPRESSION PROFILES

Abstract

Breast cancer remains one of theleading causes of cancer-related mortality worldwide, highlightingthe need for accurate diagnostic approaches and improved understanding of its molecularmechanisms. Advances in transcriptomic technologies have enabled large-scale analysis of geneexpression profiles, providing valuable opportunities for identifying molecular biomarkers associatedwith cancer development. In this study, gene expression data were analyzed to identify significantgenes differentiating tumor and normal breast tissue samples and to evaluate the effectiveness ofmachine learning models for breast cancer classification. The dataset consisted of 590 samples,including 529 tumor and 61 normal samples, with expression values measured for 17,814 genes.Differential gene expression analysis using a two-sample t-test was performed to identify informativegenes, and the top 500 statistically significant genes were selected as predictive features. Threemachine learning modelsLogistic Regression, Support Vector Machine (SVM), and RandomForestwere developed to classify tumor and normal samples based on the selected gene expression features.The dataset was divided into training and testing subsets using stratified sampling, and modelperformance was evaluated using accuracy and receiveroperating characteristic area under thecurve (ROC–AUC). The results demonstrated strong classification performance, with LogisticRegression and SVM achieving an accuracy of 97.46% and an ROC–AUC of 0.997, while RandomForest achieved an accuracy of 96.61% and an ROC–AUC of 0.994. These findings highlight thepotential of combining gene expression analysis with machine learning techniques for breast cancerclassification and biomarker discovery.