Novel feature ranking criteria for interval valued feature selection

Abstract

In this paper, novel feature ranking criteria suitable for supervised interval valued data are introduced. The ranking criterion basically used to rank the features based on their relevancy prior to feature selection for pattern classification. In our work, initially, a vertex transformation approach is applied on interval valued data to obtain with a crisp type data. Then, the proposed feature ranking criterion is applied on the vertex interval data to rank the features based on their relevancy. This followed by the selection of top k ranked features from the given d set of interval features. Thus the obtained feature subset is evaluated using suitable learning algorithm. The efficacy of the proposed ranking criteria is validated using three benchmarking interval valued datasets and two symbolic classifiers. Finally, a comparative analysis is given to uphold the superiority of the proposed model in terms of classification accuracy.