On the evaluation of unsupervised outlier detection: measures, datasets, and an empirical study