SVM-ERFE: An Enhanced Method for Identification of Metastatic and Relapsing Nature of Breast Cancer

Abstract

Classifying breast cancer into metastatic non-metastatic is crucial for a women that undergoing for surgery, because of the heavy treatment associated with metastatic cancers. With the advancement of micro array technology, identifying gene signatures from large genes dataset becomes an efficient approach to predict or classify cancerous genes. We proposed an algorithm based on support vector machine using enhanced recursive feature extraction mechanism (SVM_ ERFE). The proposed algorithm identifies most significant genes signatures and thus prognosis of metastatic/relapsing nature among breast cancer patients. In order to enhance the accuracy, we kept allow to varying the number of genes after first filtering. Several experiments has performed over metastatic-295 breast cancer patient's datasets and relaps-GSE2034 dataset, and achieved the best accuracy 100%±0.0 for 158 and 307 gene signatures after first filtering respectively. Further, the model has validated using 10-fold cross validation using evaluation metrics accuracy, TPR, TNR. Moreover, proposed model can increase the dortor's confidence into the results.