Outliers Detection in Multi-label Datasets

Abstract

In many knowledge discovery applications, finding outliers, i.e. objects that behave in an unexpected way or have abnormal properties, is more interesting than finding inliers in a dataset. Outlier detection is important for many applications, including those related to intrusion detection, credit card fraud, and criminal activity in e-commerce. Several methods of outlier detection have been proposed, and even many of them from the perspective of Rough Set Theory, but at the moment none of them is specifically intended for multi-label datasets. In this paper, we propose a method that measures the degree of anomaly of an object in a multi-label dataset. This score or measure quantifies the degree of irregularity of an object with respect to the dataset. In addition, a method for generating anomalies in this type of datasets is proposed. From these synthetic datasets, the efficacy of the proposed method is proved. The results show the superiority of our proposal over other methods in the literature adapted to multi-label problems.