Utilization of filter feature selection with support vector machine for tumours classification

Abstract

Due to rapid technology advancement, machine learning has been widely used for solving cancer classification problem. Classification performance is highly depending on the quality of input features. With an explosive increase number of features of high dimensional data, the occurrence of ambiguous samples and data redundancy directly leads to poor classification accuracy. Therefore, this paper presents a utilization of filter feature selection using four filter methods such as Information Gain, Gain Ratio, Chi-Squared and Relief-F by performing attribute rankings to remove the irrelevant and redundant features and evaluate the significance and correlation of input data. Then, the classification will be performed using Support Vector Machine (SVM) to measure the accuracy performance based on the number of selected features. The performance measurement will be validated on standard Breast Cancer datasets consisting of 286 instances obtained from the UCI repository. Evaluation metrics such as accuracy, sensitivity, specificity and Area under Receiver Operating Characteristic Curve (AUC) will be used to assess the performance of the SVM classifier using four different filter methods. Experimental result shows that Gain ratio improves the accuracy of SVM classification compared to Information Gain, Chi-Squared and Relief-F in classifying breast cancer data with only small number of features selected.