Breast Cancer Clasification Based on Gene Expression Data Using Singular Value Decompotition and Support Vector Machine

Abstract

Breast cancer gene expression data are typically high-dimensional, with the number of genes far exceeding the number of samples, which increases computational complexity and can reduce classification performance. This study aims to apply Truncated Singular Value Decomposition (TSVD) as a dimension reduction method and to evaluate the effect of rank selection on the performance of Support Vector Machines (SVMs) in classifying six breast cancer subtypes. The study employs a quantitative experimental approach using the GSE45827 dataset from the Curated Microarray Database (CuMiDa), which consists of 151 samples and 54,675 genes. The rank values were determined based on the threshold for retained information, greater than 1% and greater than 0.5%, resulting in rank-18 and rank-59. SVM models with linear and Radial Basis Function (RBF) kernels were evaluated using the F1-score and training time. The results show that the combination of TSVD with rank-18 and a linear kernel delivers the best performance, with an F1-score of 94.43% and a training time of 0.0054 seconds, outperforming the model without dimensionality reduction, which achieved an F1-score of 90.17% and a training time of 5.9684 seconds. These findings demonstrate that TSVD is capable of improving both computational efficiency and SVM classification performance on high-dimensional gene expression data.