I am doing a comparative analysis of outlier detection algorithm. I am comparing using ROC AUC. But for the datasets I am using - German fraud dataset, Breast cancer dataset for every algorithm and for most values of the parameters I am not getting AUC value greater than 0.5. What should Ido?
http://arxiv.org/ftp/arxiv/papers/1503/1503.03208.pdf
In it they compare three different clustering algorithms for outlier detection and overlap them to detect fraud more robustly.