Department of Logistics

Permanent URI for this community

Now showing 1 - 1 of 1

A comparison of methodologies with minimal hyper-parameter tuning for anomaly detection on transactional data
(Stellenbosch : Stellenbosch University, 2023-03) Alderslade, James William; Visagie, Stephan Esterhuyse; Stellenbosch University. Faculty of Economic and Management Sciences. Dept. of Logistics.
ENGLISH SUMMARY: In the information technology era where data has become as valuable as gold, being able to process and use it effectively is vital. Many large industries ranging from healthcare and insurance to banking utilise some form of transactional data in order to track and process many of the services and products which they offer to their clients. These processes are moving too fast for any human to keep up, due to the ever increasing rates of service and convenience which are core drivers in this technological era. Under these circumstances, anomaly detection is required. Fraud detection is the process whereby these transactions are identified and so are the items which could be related to any fraudulent, wasteful or abusive behaviour. A variety of anomaly detection methodologies are investigated to illustrate and compare their ability to quickly and accurately detect anomalous transactions without being given a set of rules. The methods used range from traditional statistical methods, through to machine learning and deep learning. Isolation forest performs best on the evaluation criteria used in this study.