Instance-Based Classification Through Hypothesis Testing

Classification is a fundamental problem in machine learning and data mining. During the past decades, numerous classification methods have been presented based on different principles. However, most existing classifiers cast as an optimization do not address issue of statistical significance. In this paper, we formulate binary two-sample testing problem. More precisely, our model generic framework that composed two steps. first step, distance between test instance each training calculated to derive sets. second performed under null hypothesis sets distances are drawn from same cumulative distribution. After these steps, p-values for assigned class associated with smaller p-value. Essentially, method can be regarded instance-based classifier testing. The experimental results 38 real show able achieve level performance several classic has significantly better than testing-based classifiers. Furthermore, handle outlying instances control false discovery rate framework.