Graph Based Business Process Anomaly Detection with Edge Feature Reconstruction and Advanced Linear Networks

Business Process Management (BPM) as an inter-disciplinary field between Managerial Sciences and Computer Science is a subject ever-increasing in importance. This holds more and more true as the business landscape becomes faster and more complex each passing day. Given the management of a businesses...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:International Congress on Human-Computer Interaction, Optimization and Robotic Applications (Online) s. 1 - 6
Hlavní autori: Ayaz, Teoman Berkay, Cevik, Rabia, Ozcan, Alper, Akbulut, Akhan
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 23.05.2025
Predmet:
Anomaly detection
Architecture
Autoencoder
Autoencoders
Business
Business Process
Computer architecture
Decoding
Graph Neural Network
Hybrid power systems
Image edge detection
Machine Learning
Testing
Transformers
Unsupervised
ISSN:2996-4393
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Business Process Management (BPM) as an inter-disciplinary field between Managerial Sciences and Computer Science is a subject ever-increasing in importance. This holds more and more true as the business landscape becomes faster and more complex each passing day. Given the management of a businesses operational activities is essential to maintain a healthy lifecycle, the early detection of these inefficiencies and potentially malicious activity becomes more and more crucial. As these deviations can significantly impact a businesses lifecycle, anomaly detection solutions in this domain is that much more lucrative. The pursuit of detecting these deviations gave rise to the field of Business Process Anomaly Detection. Building upon previous research, our study focuses on constructing an advanced Graph Autoencoder (GAE) architecture using various graph convolutional operators, and boost the performance further with advanced linear networks. By comprehensively evaluating 3 distinct encoder architectures and 4 distinct decoder selections, our study comprehensively evaluates the possible ways to combine various encoders and decoders on 6 distinct datasets. The empirical results show a wide range of results with varying trends between different encoder-decoder combinations, ranging from 0.674 to 0.219 F1-score in anomaly detection performance.
ISSN:2996-4393
DOI:10.1109/ICHORA65333.2025.11017131