Return to Projects

ARC Discovery on Deviance Diagnosis and Prediction

Project title: “Learning from the Good and the Bad: Diagnosis and Prediction of Business Process Deviances”

This project aims to develop an innovative approach based on process execution semantics, to analyse event data logged by IT systems in order to diagnose and predict business process deviance. Anticipated outcomes include novel business intelligence algorithms producing deviance diagnostics, predictions and recommendations and exposing results via interactive visual analytics. The outcomes are expected to aid process workers in steering business operations towards consistent and compliant outcomes and higher performance, and assist analysts and auditors to explain deviant operations. This should significantly benefit industries such as healthcare, insurance, retail and the government where compliance and integrity management are imperative.

Work Packages

Work Package I: Formal representation of process execution semantics from event logs

This package will establish the foundational notions to develop the algorithms for deviance diagnosis and prediction. These include a formal definition of compact and lossless event log representation based on process execution semantics, proofs of behavioral  losslessness and proof of compression guarantees, an algorithm to efficiently construct this representation from event logs and proofs of correctness and termination, and extensions of this representation to temporal and non-temporal process aspects.

Work Package II: Design, operationalization and evaluation of offline algorithms for deviance diagnosis

This package will build a range of offline algorithms for deviance diagnosis (incl. comparison via bisimulation and realignment of states), complemented by techniques to rank the diagnostics based on their interestingness, and verbalise them. Finally, this package will evaluate the accuracy, scalability and robustness of the designed algorithms using synthetic and real-life logs.

Work Package III: Design, operationalization and evaluation of online algorithms for deviance prediction

This package, complementary to WP2, will build a utility measure to rank process execution features for early prediction. Next, it will develop a range of online algorithms for deviance prediction (incl. probability estimation, explanation, recommendations and incremental update of the event log representation). Finally, this package will evaluate the accuracy, scalability and robustness of the designed algorithms using synthetic and real-life logs.

Work Package IV: Development of deviance management system and user evaluation

All the algorithms developed in this project will be integrated into Apromore in the form of plugins. These plugins will take as input event logs, streams of events capturing running process executions, and one or more process-based key performance indicators (KPIs), to produce diagnostics for business analysts/auditors (deviance diagnosis), as well as predictions and recommendations for process workers and managers (predictive deviance monitoring). KPIs will be encoded in the form of business/compliance rules and performance thresholds.

Sponsors

The project is co-funded by the Australian Research Council, The University of Melbourne, Eindhoven University of Technology and University of Tartu.

 

     TU/e  University of Tartu

 

Budget

Australian Research Council: AUD$378,000

  • 2018: $126,000
  • 2019: $120,000
  • 2020: $132,000

Participants

The University of Melbourne

Eindhoven University of Technology

University of Tartu