Modified ML-kNN and Rank SVM for Multi-label Pattern Classification

Abstract

To develop an efficient multi-label classifier is the main objective of this paper. In multi-label learning tasks such as classification, each example is associated with a set of labels, and the task is to predict the label set whose size is unknown apriory for each unseen example. In a realistic scenario each object or entity belongs to a multi-label category. Multi-Label k-Nearest Neighbor (ML-kNN), Rank-SVM (Ranking Support Vector Machine) are two popular techniques used for multi-label pattern classification. ML-kNN is a multi-label version of standard kNN and Rank SVM is a multi-label extension of standard SVM. The main aim of this work is to enhance the performance of these methods. Multi-label classifiers generally consider ranking loss, Hamming loss, one error, average precision and coverage as a performance metrics.