Context-aware Predictive Process Analytics (CoPPA)

流程分析现在是一种广泛使用的业务应用程序。二十年来对技术和算法的基础研究为过程数据的真实世界后处理分析奠定了基础。赛隆尼斯SE是市场领先的供应商，西门子是过程分析软件的最大用户，两家公司都位于德国。本提案中描述的研究涉及流程分析的下一个逻辑级别：预测未来流程实例的行为。当前用于预测过程分析的方法的预测能力几乎是随机的，其中大多数方法完全基于过程事件日志数据。因此，上下文感知预测性流程分析(COPPA)探索将流程上下文数据包括在对未来流程行为的预测中的机会。该项目建立在两位提出者之前的研究基础上，该研究发表在《管理信息系统季刊》上，并通过工业合作伙伴提供的数据提供支持。在此基础上，我们利用动态贝叶斯网络(DBN)的概率模型设计了一种基于上下文的预测分析技术。我们将上下文理解为与业务流程相关的每一个方面，超越了活动的纯粹名称。因此，上下文包括例如活动的输入和输出数据、执行活动的人、以及表征过程环境的传感器数据。上下文可以显著影响流程的行为，因此我们可以预期，了解流程实例中出现了什么上下文可以提高预测性流程监控方法的预测精度(例如，贷款审批流程中的贷款金额可能会对流程中做出的后续决策产生相当大的影响，从而影响其行为)。在设想的COPPA项目结束时，我们希望提供四个可交付成果：新的预测性流程分析模型和训练算法，对运行中的流程实例进行预测的新算法，新的预测性流程分析可视化概念，以及可用于设置、运行并评估预测性流程分析场景。所有这些结果都将伴随着广泛的评估指标集，因此我们将为与流程分析相关的科学知识体系做出多方面的贡献。所有创建的构件都将在开放源码软件许可证下发布，并将向公众开放。

Process analytics is now a widely used business application. Two decades of fundamental research on techniques and algorithms set the stage for real-world ex-post analysis of process data. Both Celonis SE, the market-leading vendor, and Siemens, the largest user ofprocess analytics software are located in Germany. The research described in this proposal deals with the next logical level of process analytics: predicting the behavior of future process instances. The predictive power of current approaches to predictive process analytics, most of which operate exclusively on process-event log data, is close to random. Therefore, context-aware predictive process analytics (CoPPA) explores opportunities to include process-context data in predictions of future process behavior. The project builds upon prior research of the two proposers that was published in MIS Quarterly and is support through data provision by industrial partners. We design a context-based predictive analytics technique using probabilistic models from the class of Dynamic Bayesian Networks (DBN) on that basis. We understand context as every aspect related to a business process that goes beyond the sheer name of an activity. Hence, context includes, for instance, input and output data of an activity, people performing an activity, and sensor data characterizing the environment of a process. Context can significantly influence the behavior of a process, so we can expect that being aware of what context has been present in a process instance can increase the prediction accuracy of a predictive process monitoring approach (e.g., the amount of a loan in a loan approval process may have a considerable influence onto the subsequent decisions made in the process and thus influences its behavior).At the end of the envisioned CoPPA project, we expect to provide four deliverables: novel predictive process analytics models and training algorithms, novel algorithms that make predictions on running process instances, novel predictive process analytics visualization concepts, and a scientific workflow environment that can be used to set up, run, and evaluate predictive process analytics scenarios. All of these outcomes will come with extensive sets of evaluation metrics, so we will make multiple contributions to the scientific body of knowledge related to process analytics. All created artifact will be published under an open-source software license and they will be made publicly accessible.