A Probabilistic Combination Approach to Improve Outlier Detection

In this paper we propose a probabilistic approach to combine the results from multiple outlier detection algorithms. In our approach, we first estimate an outlier score vector for each data object. Each element of the estimated vectors corresponds to an outlier score produced by a specific outlier detection algorithm. We then use the multivariate beta mixture model to cluster the outlier score vectors into several components so that the component that corresponds to the outliers can be identified. We illustrate the suitability of our proposal through an empirical study that uses both artificial and real-life data sets. Our results show that the proposed approach enhances the results of the combined outlier detection algorithms, and avoids their pitfalls.