Deep Learning Based on TensorFlow and Keras for Predictive Monitoring of Business Process Execution Delays

In order to enhance their performance and responsiveness, organizations must identify, manage, and monitor all business processes that involve crucial knowledge. This can be achieved through a multidisciplinary approach that combines Knowledge Management, Business Process Management, and Process Mining. To achieve these objectives, the implementation of an automated computer system for managing business processes has become paramount. In this context, we adopt the CRISP-DM approach to propose a new method called BPEDPM (Business Process Event Data Predictive Monitoring) for predictive process monitoring. This method leverages process mining techniques to exploit the execution data from a Business Process Management System (BPMS) workflow engine. Particularly, in the modeling phase, we apply deep learning based on the TensorFlow and Keras tools. To demonstrate the applicability of the BPEDPM method, we have developed an intelligent system named iBPMS4PED to predict the execution times of business processes. The research focuses on the incoming mail management process within a group health insurance context.