Explainable Predictive Process Monitoring: Evaluation Metrics and Guidelines for Process Outcome Prediction

Abstract

Recently, a shift has been made in the field of Outcome-Oriented Predictive Process Monitoring (OOPPM) to use models from the eXplainable Artificial Intelligence paradigm, however the evaluation still occurs mainly through performancebased metrics not accounting for the implications and lack of actionability of the explanations. In this paper, we define explainability by the interpretability of the explanations (through the widely-used XAI properties parsimony and functional complexity) and the faithfulness of the explainability model (through monotonicity and level of disagreement). The introduced properties are analysed along the event, case, and control flow perspective that are typical of a process-based analysis. This allows to quantitatively compare, inter alia, inherently created explanations (e.g., logistic regression coefficients) with post-hoc explanations (e.g., Shapley values). Moreover, this paper contributes a guideline named X-MOP to practitioners to select the appropriate model based on the event log specifications and the task at hand, by providing insight into how the varying preprocessing, model complexity and post-hoc explainability techniques typical in OOPPM influence the explainability of the model. To this end, we benchmark seven classifiers on thirteen real-life events logs.

Assessing the Robustness in Predictive Process Monitoring through Adversarial Attacks

“…In [16], the authors demonstrate the inability of Long Short-Term Memory (LSTM) models to generalize process model behaviour without careful measurement and evaluation. In previous work [7], the faithfulness of post-hoc explanations was shown to be compromised. The authors of [17] introduced an approach to train robust and generalizable predictive models that can handle spurious data correlations.…”

“…Recent works have already issued the lack of reliability of deep learning models in the context of predictive process monitoring [7], [16], [17]. In [16], the authors demonstrate the inability of Long Short-Term Memory (LSTM) models to generalize process model behaviour without careful measurement and evaluation.…”

“…Nonetheless, the inability to comprehend the decision-making of these models prohibits a successful adoption thereof. This has given rise to eXplainable AI (XAI) into the field of OOPPM [5]- [7] which relies on either inherently transparent models, or opaque models with post-hoc explanation techniques such as SHapley Additive exPlanations (SHAP) values [7]. A successful adoption of these models is conditional upon the trust that the decision-makers have in the predictions and the explanations, and whether they conform to the process domain knowledge [8], [9].…”