FINDING FEATURES FOR THE PREDICTION OF BREAST CANCER USING THE K NEAREST NEIGHBORHOOD AND SUPPORT MACHINES METHODS

Abstract

Breast cancer, the most common type of cancer among women, affects millions of women every year. The aim of this study is to find properties for estimating breast cancer using two basic classifier methods and to compare the classifier performances. In the study, a ready data set including age, body mass index (BMI), glucose, resistin, insulin, homeostasis model assessment for insulin resistance (HOMA-IR), monocyte chemoattractant protein 1 (MCP1), leptin hormone and adiponectin hormone were used. 10 fold cross verification (10 fold) method was used while selecting the training and test data. Classification performance of SVM from classifiers was better than KNN. It was seen that the data set with 5 clinical features showed the best classification performance with 85.3% classification accuracy and 89.1% specificity (the rate of detecting those with cancer) values. It has been determined that the most suitable features for breast cancer prediction are age, BMI, resistin, glucose and adiponectin parameters.