Unsupervised Learning Methods for Anomaly Detection and Log Quality Improvement Using Process Event Log

  • M. Vijayakamal, D. Vasumathi


With Service Oriented Architecture (SOA) prevailing, enterprise applications of the real world became very complex in size and operations. This has led to several thousands of processes involved. Manual inspection of processes became very tedious and impractical exercise. Process mining came into existence to deal with extraction or discovery of business processes, finding outliers and anomalies besides process enhancement. Out of these aspects, anomaly detection and process enhancement are given high importance. The existing methods in this area have limitations in terms of accurately identifying anomalies and enhancing the processes. To overcome this problem, a comprehensive framework is proposed for detecting anomalies from business process event logs and rectify anomalies for process enhancement. The Simple Auto Encoder based Anomaly Detection (SAE-AD) algorithm is proposed to achieve this. However, it is deterministic in nature while making prediction decisions. A probabilistic algorithm called Probabilistic Auto Encoder based Anomaly Detection (PAE-AD) is proposed to detect anomalies and enhance processes involved. Both algorithms are based on Artificial Neural Networks (ANNs) with unsupervised learning. Empirical study is made with Business Process Intelligence (BPI) challenge datasets especially targeted for process mining research. The empirical results revealed that the proposed framework improves performance in anomaly detection and process enhancement.

Keywords:–Business process event log, process mining, Business Process Intelligence (BPI), anomaly detection


How to Cite
D. Vasumathi, M. V. (2020). Unsupervised Learning Methods for Anomaly Detection and Log Quality Improvement Using Process Event Log. International Journal of Advanced Science and Technology, 29(1), 1109 - 1125. Retrieved from http://sersc.org/journals/index.php/IJAST/article/view/3603