Causal Graph: Interpretation of Causal Relationships in Temporary Deviations of Business Processes

Process deviations can be difficult and costly to identify. Therefore, it's imperative for organizations to detect temporary deviations during execution and understand their causal relationships. This enables decision-makers to implement targeted corrective actions. This article presents a method to construct a causal graph for the analysis of process variants, combining techniques from process mining, unsupervised machine learning, and causal discovery and inference. This graph is not susceptible to Simpson's paradox, where aggregating the feature space can lead to incorrect interpretations of causal effects. The technique has been initially validated with a well-known event log, namely a loan application process taken from the BPI Challenge 2017 and containing 16,299 records. This pilot run successfully identified the causal variables and their direction. Wider use of this approach will allow organizations to interpret and estimate the causal effect of an action plan on process variants with temporary deviations.