Finding Anomalies in Millions of Records: A Statistical Approach vs Machine Learning Algorithms