This repository contains experiments described in the paper Data-Driven Decision Support for Business Processes: Causal Reasoning and Discovery by Ali J. Alaee, Mathias Weidlich, and Arik Senderovich.

DataGenerator_mainprocess.ipynb
This notebook generates a synthetic dataset for the main process example provided in the paper. Based on the causal graph in Fig. 2, it creates 10,000 samples, converts them into a trace log, and finally into an event log. The resulting .csv file can be used with the Bayesys causal discovery tool to identify the underlying causal graph. Additionally, the event log can be used with the Inductive Miner to discover the process map.

BPI2012_TabularData_Extraction.ipynb
This notebook extracts a tabular dataset of decisions and contextual variables from the real-world BPI Challenge 2012 dataset. The output .csv file can be used with the Bayesys causal discovery tool to identify the underlying causal graph.

ATE_calculation.ipynb
This notebook facilitates the calculation of Average Treatment Effects (ATE) and Conditional Average Treatment Effects (CATE) using do-calculus. It requires an input causal graph and a .csv dataset.

Various types of decisions influence the execution of a business process, e.g., in terms of control-flow and resource assignments. Data recorded during process execution can be used to identify which decisions are informed by data and by previous decisions, to predict their outcome, and to guide interventions as part of a what-if analysis. The latter requires causal models that explain decisions. Yet, existing causal techniques for business processes are limited: they focus on control-flow decisions only, ignore confounding variables, and use ad-hoc methods to resolve causal conflicts. In this paper, we fill this gap, by introducing a causal decision modeling framework that incorporates data variables, which uncover confounding effects, and captures resource decisions. Moreover, we provide a process-aware causal discovery algorithm, based on the notion of temporal tiers, that takes process precedence into account, without the need for heuristic conflict resolution between process discovery and causal discovery. We demonstrate the effectiveness of our approach through experiments with a synthetically generated dataset, and show a proof-of-concept implementation on a real-world dataset.