Comparison of Different Classifiers for Prediction of Breast Cancer Metastasis in Microarray Analysis

Abstract

Background: In this research, we investigated the performance of some different classifiers for prediction of metastasis in breast cancer. Methods: We used the DNA microarrays of primary breast tumors of 78 young patients. Among these patients, 34 had developed distant metastases within 5 years (poor prognosis group) and 44 formed good prognosis group. For analysis, we applied three different classifiers including support vector machine (SVM), stepwise linear discriminant analysis (SWLDA) and K-nearest neighbors (KNN) classifier. Each of these classifiers used 231 selected genes as an input feature vector and their performances were estimated via using leave one out (LOO) method to classify patients into two groups namely, good and poor prognosis. Findings: The best results were obtained by support vector machine with linear kernel. This classifier achieved a sensitivity and specificity of 84% and 82%, respectively, for metastasis prediction. Conclusion: Our findings provide a strategy to specify patients who would benefit from adjuvant therapy.

Keywords

• Microarrays
• Prediction of breast cancer
• Support vector machine (SVM)
• Stepwise linear discriminant analysis (SWLDA)
• k-nearest neighbors (KNN) classifiers