This dataset contains 9 observations for each of 683 breast cancer biopsies.
The biopsies were classified by experts into two categories: Benign (444) and Malignant (239).
Selecting the 9 observations for all biopsies, but (of course) leaving out the expert classification, we can test Anomaly Server’s ability to detect an adversarial minority in the dataset.
The graph below shows the anomaly score, as calculated by Anomaly Server, versus the accumulated percentage of benign and malignant biopsies, respectively.
As is clearly visible from the graph, the lower the anomaly score, the greater chance that a biopsy is benign.
In this analysis, the samples with an anomaly score less than 0.5 contains 95% of all benign biopsies, but only 2% of the malignant.
Notice, that because Anomaly Server uses randomised sampling of the data, the results may vary slightly from calculation to calculation.