A Review of Unsupervised Machine Learning Frameworks for Anomaly Detection in Industrial Applications

Abstract

Unsupervised learning, also known as unsupervised machine learning, analyzes and clusters unlabeled data utilizing machine learning techniques. Without human input, these algorithms discover patterns or groupings in the data. In the domain of abuse and network intrusion detection, interesting objects are often short bursts of activity rather than rare objects. Anomaly detection is a difficult task that requires familiarity and a good understanding of the data and the pattern does not correspond to the common statistical definition of an outlier as an odd item. The traditional algorithms need data preparations while unsupervised algorithms can be prepared so that they can handle the data in war format. Anomaly detection, sometimes referred to as outlier analysis is a data mining procedure that detects events, data points, and observations that deviates from the expected behavior of a dataset. The unsupervised machine learning approaches have shown potential in static data modeling applications such as computer vision, and their use in anomaly detection is gaining attention. A typical data might reveal critical flaws, such as a software defect, or prospective possibilities, such as a shift in consumer behavior. Currently, academic literature does not really cover the topic of unsupervised machine learning techniques for anomaly detection. This paper provides an overview of the current deep learning and unsupervised machine learning techniques for anomaly detection and discuss the fundamental challenges in anomaly detection.