Categorical anomaly detection in heterogeneous data using minimum description length clustering