Improvised mRMR feature selection for predicting breast cancer

Abstract

The focus of the proposed method is to provide a solution to the problem of predicting the presence of breast cancer for the data in the UCI Repository. The strong ideology of the proposed method is to predict the presence of cancerous information based on the details of parameters from UCI Repository. The feature selection of proposed method tunes certain parameters to select only few features which are most essential and relevant and far away from the redundant information. The output of feature selection algorithm is given to the SVM classifier with various parameters to train and test in the ratio of 90:10, where 90% of information is considered as Training data with proposed method and the rest 10% of data is considered as a Test the data. The proposed method has included the improvement in mRMR feature selection by tuning the parameters of features with respect to feature set. Thus, the proposed statistical approach has yielded a good result of 98.3% accuracy during the testing phase against the training phase over the UCI Wisconsin data repository.