Location

Hilton Hawaiian Village, Honolulu, Hawaii

Event Website

https://hicss.hawaii.edu/

Start Date

3-1-2024 12:00 AM

End Date

6-1-2024 12:00 AM

Description

The performance of a service process can be improved by the early anticipation of future behavior, such as predicting the next activity using predictive business process monitoring (PBPM). Recent PBPM techniques are based on deep neural networks (DNNs) and consider the process context to create accurate predictions. To provide explainability of these predictions, model-agnostic explainable AI (XAI) methods, for example, SHAP, can be used. However, creating these explanations is time-consuming and, therefore, not applicable to service processes where customers are involved. In this paper, we propose a context-aware DNN-based technique to efficiently create meaningful explanations of next activity predictions using layer-wise relevance propagation. We evaluate the predictive quality and the explanation creation time, using three real-life service event logs. Further, we demonstrate its visual output, highlighting its utility for end-users.

Share

COinS
 
Jan 3rd, 12:00 AM Jan 6th, 12:00 AM

Context-aware Explanations of Accurate Predictions in Service Processes

Hilton Hawaiian Village, Honolulu, Hawaii

The performance of a service process can be improved by the early anticipation of future behavior, such as predicting the next activity using predictive business process monitoring (PBPM). Recent PBPM techniques are based on deep neural networks (DNNs) and consider the process context to create accurate predictions. To provide explainability of these predictions, model-agnostic explainable AI (XAI) methods, for example, SHAP, can be used. However, creating these explanations is time-consuming and, therefore, not applicable to service processes where customers are involved. In this paper, we propose a context-aware DNN-based technique to efficiently create meaningful explanations of next activity predictions using layer-wise relevance propagation. We evaluate the predictive quality and the explanation creation time, using three real-life service event logs. Further, we demonstrate its visual output, highlighting its utility for end-users.

https://aisel.aisnet.org/hicss-57/da/service_analytics/3