Whetting the SWORD: Detecting Workarounds by Using Active Learning and Logistic Regression
| dc.contributor.author | Van Der Waal, Wouter | |
| dc.contributor.author | Weerd, Inge | |
| dc.contributor.author | Haitjema, Saskia | |
| dc.contributor.author | Kappen, Teus | |
| dc.contributor.author | Alexander Reijers, Hajo | |
| dc.date.accessioned | 2023-12-26T18:41:55Z | |
| dc.date.available | 2023-12-26T18:41:55Z | |
| dc.date.issued | 2024-01-03 | |
| dc.identifier.doi | 10.24251/HICSS.2023.445 | |
| dc.identifier.isbn | 978-0-9981331-7-1 | |
| dc.identifier.other | dc2152b0-9510-4704-be67-cb5fbe0be80f | |
| dc.identifier.uri | https://hdl.handle.net/10125/106828 | |
| dc.language.iso | eng | |
| dc.relation.ispartof | Proceedings of the 57th Hawaii International Conference on System Sciences | |
| dc.rights | Attribution-NonCommercial-NoDerivatives 4.0 International | |
| dc.rights.uri | https://creativecommons.org/licenses/by-nc-nd/4.0/ | |
| dc.subject | Process Mining in Healthcare | |
| dc.subject | active learning | |
| dc.subject | event log | |
| dc.subject | logistic regression | |
| dc.subject | process mining | |
| dc.subject | workarounds | |
| dc.title | Whetting the SWORD: Detecting Workarounds by Using Active Learning and Logistic Regression | |
| dc.type | Conference Paper | |
| dc.type.dcmi | Text | |
| dcterms.abstract | In many organizations, especially in healthcare, workers may work around prescribed procedures. Detecting these workarounds can give insights into difficulties concerning the procedures, which in turn can be used to improve them. Previous studies have shown that workarounds may be discovered from an event log using a set of predefined patterns such as the duration of a trace or the number of resources involved in one. However, domain experts may find it difficult to evaluate and monitor results if there are multiple patterns that indicate workarounds. Training a model that merges the features is often difficult because there are no available datasets covering workarounds. Labeling traces generally requires a lot of time from domain experts. In addition, this would have to be repeated for every new domain, company, or even department since the types of workarounds that occur may differ strongly between them. In this work, we propose to combine the features using a Logistic Regression model and train through Active Learning. In a case study at a hospital, we find that after training the model on only 10 to 15 traces, it stabilizes with an approximate F1 score of .75. This shows that we create and train a model that can detect workarounds well without requiring a large amount of labeled data or a lot of time from a domain expert. | |
| dcterms.extent | 10 pages | |
| prism.startingpage | 3687 |
Files
Original bundle
1 - 1 of 1