Open Access System for Information Sharing

Login Library

 

Thesis
Cited 0 time in webofscience Cited 0 time in scopus
Metadata Downloads

Multi-Perspective Anomaly Detection in Process Mining

Title
Multi-Perspective Anomaly Detection in Process Mining
Authors
김도현
Date Issued
2018
Publisher
포항공과대학교
Abstract
Many enterprises put lots of effort in improving their business processes to achieve higher level of productivity and satisfaction of employees. One of the methods to improve business processes is to detect deviating behaviors, also known as anomalies, and to remove them from the processes. Until now, most of the researches focus on anomaly detection from control-flow perspective, finding low frequency process patterns and patterns that deviate from standard, reference process models. However, event logs accumulated through business processes are not only limited to control-flows; it also contains information about the time when an event has occurred and the resource who has performed the event. Therefore, it is also necessary to look at business processes from time and resource perspectives to rich the quality of anomaly detection methods. Therefore, in this paper, we propose a new, novel algorithm to detect anomalies from multi-perspectives. The algorithm consists of five simple steps, which are preprocessing, computation of performance measures, rule generation from multi-perspectives, rule matching, and computation of anomaly scores, to detect whether a case is normal or anomaly. The algorithm is validated with both artificial and real-life event logs to see how well the proposing algorithm works on different environment. It is also compared with an existing anomaly detection technique to show its novelty.
작업자의 만족도와 작업에 대한 높은 생산성을 성취하기 위해 많은 기업은 비즈니스 프로세스를 개선하려고 노력한다. 개선의 한 방법으로 프로세스 내 발생하는 이상치를 탐지하는 방법이 있다. 현재까지 진행된 대부분의 연구는 작업 간 순서만을 고려한 이상 탐지에 대해서 진행되어 왔다. 예시로, 기준이 되는 프로세스 모델과 차이를 발견하거나 빈번하게 발생하지 않는 프로세스 패턴을 감지하는 것이 있다. 하지만 보다 효과적인 이상 탐지를 하기 위해서는 비즈니스 프로세스를 시간적 관점과 작업자 관점에서도 바라볼 필요가 있다. 따라서 본 연구에서는 작업 간 순서, 시간, 그리고 작업자 관점에서 이상 탐지를 할 수 있는 방법을 제안한다. 본 방법을 가상 데이터와 실 데이터에 적용하여 다양한 환경에서 본 방법이 잘 작동하는지 살펴본다.
URI
http://postech.dcollection.net/common/orgView/200000106244
https://oasis.postech.ac.kr/handle/2014.oak/92844
Article Type
Thesis
Files in This Item:
There are no files associated with this item.

qr_code

  • mendeley

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Views & Downloads

Browse