Jana-Rebecca Rehse

Predicting process behaviour using deep learning

Abstract Predicting business process behaviour is an important aspect of business process management. Motivated by research in natural language processing, this paper describes an application of deep learning with recurrent neural networks to the problem of predicting the next event in a business process. This is both a novel method in process prediction, which has largely relied on explicit process models, and also a novel application of deep learning methods. The approach is evaluated on two real datasets and our results surpass the state-of-the-art in prediction precision.

A Deep Learning Approach for Predicting Process Behaviour at Runtime

Predicting the final state of a running process, the remaining time to completion or the next activity of a running process are important aspects of runtime process management. Runtime management requires the ability to identify processes that are at risk of not meeting certain criteria in order to offer case managers decision information for timely intervention. This in turn requires accurate prediction models for process outcomes and for the next process event, based on runtime information available at the prediction and decision point. In this paper, we describe an initial application of deep learning with recurrent neural networks to the problem of predicting the next process event. This is both a novel method in process prediction, which has previously relied on explicit process models in the form of Hidden Markov Models (HMM) or annotated transition systems, and also a novel application for deep learning methods.

Process Mining and the Black Swan: An Empirical Analysis of the Influence of Unobserved Behavior on the Quality of Mined Process Models

In this paper, we present the epistomological problem of induction, illustrated by the metaphor of the black swan, and its relevance for Process Mining. The quality of mined models is typically measured in terms of four dimensions, namely fitness, precision, simplicity, and generalization. Both precision and generalization rely on the definition of “unobserved behavior”, i.e. traces not contained in the log. This paper is intended to analyze the influence of unobserved behavior, the potential black swan, has on the quality of mined models. We conduct an empirical analysis to investigate the relation between a system, its observed and unobserved behavior and the mined models. The results show that the unobserved behavior, mainly determined by the nature of the unknown system, can have a significant impact on the quality assessment of mined models, hence eliciting the need to explicate and discuss the assumptions underlying the notions of unobserved behavior in more depth.

Business process management for Industry 4.0 – Three application cases in the DFKI-Smart-Lego-Factory

Abstract The advent of Industry 4.0 is expected to dramatically change the manufacturing industry as we know it today. Highly standardized, rigid manufacturing processes need to become self-organizing and decentralized. This flexibility leads to new challenges to the management of smart factories in general and production planning and control in particular. In this contribution, we illustrate how established techniques from Business Process Management (BPM) hold great potential to conquer challenges in Industry 4.0. Therefore, we show three application cases based on the DFKI-Smart-Lego-Factory, a fully automated “smart factory” built out of LEGO® bricks, which demonstrates the potentials of BPM methodology for Industry 4.0 in an innovative, yet easily accessible way. For each application case (model-based management, process mining, prediction of manufacturing processes) in a smart factory, we describe the specific challenges of Industry 4.0, how BPM can be used to address these challenges, and, their realization within the DFKI-Smart-Lego-Factory.

Process Mining Crimes – A Threat to the Validity of Process Discovery Evaluations

Given the multitude of new approaches and techniques for process mining, a thorough evaluation of new contributions has become an indispensable part of every publication. In this paper, we present a set of 20 scientifically supported “process mining crimes”, unintentional mistakes that threaten the validity of process discovery evaluations. To determine their prevalence even in high-quality publications, we perform a meta-evaluation of 21 process discovery papers published at the BPM conference. We find that none of these papers is completely crime-free, but the number of crimes and their impact on the evaluations’ validity differs considerably. Based on our list of crimes, we suggest a catalog of 13 process mining guidelines, which may contribute to avoiding process mining crimes in future evaluations. Our objective is to spark an open discussion about the necessity of valid evaluation results among both process mining researchers and practitioners.

Process Discovery from Event Stream Data in the Cloud - A Scalable, Distributed Implementation of the Flexible Heuristics Miner on the Amazon Kinesis Cloud Infrastructure

Cloud computing offers readily available, scalable infrastructure to tackle problems involving high data volume and velocity. Discovering processes from event streams, especially when the business processes execute in a cloud environment, is such a problem. Event stream data is generated rapidly with varying volume and must be processed on-the-fly, making stream processing an important use case for cloud computing. This paper describes a distributed, streaming implementation of the flexible heuristics miner on Amazon Kinesis, a cloud-based event stream infrastructure, showing how mining methods can scale effortlessly to tens of millions of events per minute.