Anomaly Detection With Multivariate K-sigma Score Using Monte Carlo

Abstract

We propose a new unsupervised anomaly detection technique based on Monte Carlo sampling. We draw samples from a Gaussian probability distribution that fits to the given data and use these samples to form a Monte Carlo estimate of an anomaly score. Many of the current anomaly detection algorithms detect anomalies without producing any anomaly scores. This may lead to confusion over the scaling of the anomaly situation as well as identifying the underlying cause for anomalies in explainable conditions. The proposed algorithm is a multi-variate unsupervised technique, which also provides anomaly scores that allows measuring the strength of outlierness of a multi-variate data point. Our experiments show that the proposed algorithm achieves better results compared to the state-of-the-art anomaly detection algorithms with clustering-based approaches and isolation forest.