Anomaly Detection on Health Data

The identification of anomalous records in medical data is an important problem with numerous applications such as detecting anomalous reading, anomalous patient health condition, health insurance fraud detection and fault detection in mechanical components. This paper compares the performances of seven state-of-the-art anomaly detection algorithms to do detect anomalies in healthcare data. Our experimental results in six datasets show that the state-of-the-art method of isolation based method iForest has a better performance overall in terms of AUC and runtime.