Performance evaluation and comparative analysis of various machine learning techniques for diagnosis of breast cancer

Breast cancer is heterogeneous and life threatening diseases among women in world wide. The aim of this paper is to analyze and investigate a novel approach based on NSST (Shearlet transform) to diagnosis the digital mammogram images. Shearlet Transform is a multidimensional version of the composite dilation wavelet transform, and is especially designed to address anisotropic and directional information at various scales. Initially, using multi scale directional representation, mammogram images are decomposed into different resolution levels with various directions from 2 to 32. In this work we investigated five machine learning algorithm, namely SVM (Support Vector Machine), Naive Bayes, KNN LDA and MLP, which are used to categorizes decomposed image as either cancerous (abnormal) or not (normal) and then again abnormal severity is further categorized as either benign images or malignant images. The evaluation of the system is carried out on the MIAS (Mammography Image Analysis Society) database. The tenfold cross-validation test is applied to validate the developed system. The performance of the five algorithms was compared to find the most suitable classifier. At the end of the study, obtained results shows that SVM is an efficient technique compares to other methods.