Robust Geodesic based Outlier Detection for Class Imbalance Problem

Abstract Outlier detection is very useful in many applications, such as fraud detection and network intrusion detection. However, some existing methods often generate incorrect identification results due to the imbalanced distribution of data points. In this paper, we present a robust geodesic-based outlier detection algorithm which simultaneously considers both global disconnectivity score and local real degree as measures of outlierness. We first construct the global disconnectivity score to incorporate suitable global characteristics of data, then we provide the local real degree to effectively consider the local characteristics of points. Thus, we can identify local outliers with higher overall connectivity but in a smaller cluster with fewer points. Experimental results obtained for a number of synthetic and real-world data sets demonstrate the effectiveness and robustness of our method. In particular, we estimate an increase in average area under curve (AUC) on ten datasets of approximately 15%, with smaller RMSD than any of the competing methods.