EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
1093. Approaches for multi-perspective trace clustering explainability
Invited abstract in session TB-27: Advances in Process Mining, stream Mathematical Optimization for XAI.
Tuesday, 10:30-12:00Room: 047 (building: 208)
Authors (first author is the speaker)
| 1. | Yannis Bertrand
|
| 2. | Jochen De Weerdt
|
| Decision Sciences and Information Management, KU Leuven | |
| 3. | EstefanÃa Serral
|
| Research Centre for Information Systems Engineering (LIRIS), KU Leuven |
Abstract
IoT devices are often present around business processes (BPs), collecting data about parameters that have an impact on the process. E.g., the pressure in a tank can be a relevant aspect of the execution of a BP.
Trace clustering (TC) is used to group similar process instances, usually based on their sequential activity patterns. TC is often used to produce models describing different process variants discovered from separate clusters. However, TC can also be applied for exploratory data analysis (EDA).
This is particularly interesting in the context of multi-perspective BPs, which are surrounded with IoT devices. Multi-perspective TC (MPTC) can be used to group together cases with similar activity sequences and sensor data. However, explaining these clusters to process users then becomes more challenging.
To shed some light in the results of MPTC, we propose to use various complementary explainability approaches:
1) Partial dependency plots: they give a visual representation of the contribution of each perspective to the TC, and help users explore their clusters.
2) Permutation feature importance: give a more quantitative measure of the impact of each perspective, on the global TC.
3) Prototypes and criticisms: Show concrete instances and actively try to challenge TC results, showing the robustness of the clusters.
Each of these approaches focuses on a particular aspect of the MPTC, and supports a better understanding of the sources of variation of the BP.
Keywords
- Artificial Intelligence
Status: accepted
Back to the list of papers