Characterisation of anomaly detection algorithms using simulated dataset for algorithm selection in application cases. (3rd June 2022)

• Record Type: Journal Article
• Authors: Divya, D.; Bhasi, M.; Kumar, M.B. Santosh
• Is Part Of: International journal of embedded systems. Volume 15:Number 2(2022)
• Page Start: 93
• Page End: 107
• Publication Date: 2022-06-03
• Subjects: anomaly detection -- data statistics -- anomaly statistics -- benchmark -- characterisation -- data generation -- simulation -- application framework -- case study -- recommendation system -- detection rate; Embedded computer systems -- Periodicals; 004.16
• Languages: English
• ISSNs: 1741-1068

Abstract

Analysing the suitability of a particular anomaly detection algorithm in a real-time scenario is a challenging task since the benchmark studies are either restricted to a specific application or limited to characterisation study of outliers from a statistical point of view. This study makes an effort to design a system that performs characterisation study of anomaly detection algorithms based on data and anomaly statistics. Comparison of different anomaly detection algorithms was done by conducting an experimental study on a generated dataset that follows various data distributions and has distinct types and percentages of anomalies. Using this simulation model an application framework is proposed which is applicable to multiple industrial domains. Validation of the system is done using the case analysis that guides the users to utilise the recommender system for their application. Output of the recommender system suggests a suitable algorithm for their specific application.