Explainability in Process Outcome Prediction: Guidelines to Obtain Interpretable and Faithful Models

Predicting the outcome of various sorts of processes has seen a strong uptake in recent years due to the advances in machine and deep learning. Although a recent shift has been made in the field of process outcome prediction to use models from the explainable artificial intelligence field, the evaluation still occurs mainly through predictive performance-based metrics, thus not accounting for the explainability, actionability, and implications of the results of the models. This paper addresses explainability through the properties interpretability and faithfulness in the field of process outcome prediction. We introduce metrics to analyse these properties along the main dimensions of (business) process data: the event, case, and control flow attributes. This allows comparing inherently created explanations with post-hoc explanations techniques, for which we benchmark seven classifiers on thirteen real-life events logs covering a range of transparent and non-transparent machine learning and deep learning models complemented with post-hoc (modelagnostic or model-dependent) explainability techniques. Next, this paper contributes a set of guidelines named X-MOP which allows selecting the appropriate model by providing insight into how the varying preprocessing, model complexity, and explainability techniques typical in process outcome prediction influence the explainability of the model.

Alexander Stevens, Johannes De Smedt

Link to paper: PDF.

Quantifying Explainability in Outcome-Oriented Predictive Process Monitoring

The growing interest in applying machine and deep learning algorithms in an Outcome-Oriented Predictive Process Monitoring (OOPPM) context has recently fuelled a shift to use models from the explainable artificial intelligence (XAI) paradigm, a field of study focused on creating explainability techniques on top of AI models in order to legitimize the predictions made. Nonetheless, most classification models are evaluated primarily on a performance level, where XAI requires striking a balance between either simple models (e.g. linear regression) or models using complex inference structures (e.g. neural networks) with post-processing to calculate feature importance. In this paper, a comprehensive overview of predictive models with varying intrinsic complexity are measured based on explainability with model-agnostic quantitative evaluation metrics. To this end, explainability is designed as a symbiosis between interpretability and faithfulness and thereby allowing to compare inherently created explanations (e.g. decision tree rules) with posthoc explainability techniques (e.g. Shapley values) on top of AI models. Moreover, two improved versions of the logistic regression model capable of capturing non-linear interactions and both inherently generating their own explanations are proposed in the OOPPM context. These models are benchmarked with two common state-of-the-art models with posthoc explanation techniques in the explainability-performance space.

Alexander Stevens, Johannes De Smedt, Jari Peeperkorn

Link to paper: here