Multivariate Conditional Outlier Detection: Identifying Unusual Input-Output Associations in Data.

We study multivariate conditional outlier detection, a special type of the conditional outlier detection problem, where data instances consist of continuous input (context) and binary output (responses) vectors. We present a novel outlier detection framework that identifies abnormal input-output associations in data using a decomposable conditional probabilistic model. Since the components of this model can vary in their quality, we combine them with the help of weights reflecting their reliability in assessment of outliers. We propose two ways of calculating the component weights: global that relies on all data and local that relies only on the instances similar to the target instance. Experimental results on data from various domains demonstrate the ability of our framework to successfully identify multivariate conditional outliers.