A Critical Review of the Applications and AI Techniques for Anomaly Detection

In the process of analysing data, outlier detection (i.e., anomaly detection or novelty identification) is often misinterpreted to an identification of rare observations, occurrence or an item, which deviates highly from enormous data and never conforms to well- defined ideologies of a normal behaviour. The samples could stimulate more suspicion of being produced from various techniques, or appear unpredictable with the remaining portion of the specific dataset. Anomaly detection draws application in different domains such as neuroscience, statistics, machine vision, medicine, financial fraud, law enforcement and cyber security. The data that has been collected from real-life applications are rapidly increasing in dimension and size. As the aspect of dimensionality keeps increasing, data items become significantly sparse, amounting to an identification of variances becoming problematic. In addition, more conventional approaches for anomaly detection cannot function in a proper manner. In this paper, we have evaluated the applications and methods of anomaly detection.