A novel unsupervised method for root cause analysis of anomalies using sparse optimization techniques

Satheesh K. Perepu, Vivek Shankar Pinnamaraju

Abstract: Anomaly detection is an imperative problem associated with emerging fields such as the Internet of Things, Telecommunication and Manufacturing Industries. Timely detection of anomalies and attribution to its sources enables to enforce preventive measures so as to maintain optimal process operation and avoid undesirable down-times. In this work, we propose an unsupervised method which enables detection, quantification and diagnosis of amplitude anomalies from multivariate data. The proposed method utilizes the ideas of sparse optimization to effectively decompose the data contributions into with and without anomalies for the purpose of anomaly detection and quantification. Further, the ideas of the directed graph are implemented on the estimated anomaly-free data to determine the root cause of anomaly to enable preventive maintenance. The efficacy of the proposed method is demonstrated on two case studies comprising of synthetic and real-time datasets.