Outlier detection from large distributed databases