Development of Statistically Modelled Feature Selection Method for Microwave Breast Cancer Detection

Abstract

Microwave technology is very promising tool for breast cancer detection. Microwave transmits and receives UWB signals. UWB signals carries information of the breast cancer. UWB signals need to be pre-processed in order to remove irrelevant and redundant features. Feature extraction and feature selection methods are mostly used to remove the unwanted features. In this paper, a statistically modelled feature selection (SMFS) method is proposed for microwave breast cancer detection. Initially, performance of different feature extraction and feature selection method are analysed using Anova test (p-value) and machine learning (SVM, DT, PNN, NB) accuracy. The best feature extraction and feature selection methods are combined and tested. Based on the performance of feature extraction and feature selection method, Combined Neighbour Component Analysis (feature selection) and Statistical features (feature extraction) are combined and tested. This method is able to achieve up to 85%. The result proves two stage methods are able to improve the accuracy compared to single stage method. Therefore, SMFS is able to detect breast cancer efficiently.