Simulation of Real-world Event Repositories for Evaluation of Data Analytics Solutions: Case of User Behavior Pattern Recovery

Due to the lack of access to the real-world event-log repositories in critical domains such as healthcare and banking, the evaluation and maintenance of data analytics algorithms has become a challenge. Generating synthetic log repositories that simulate a variety of complex real-world event-log repositories will be an effective way of producing benchmarks to evaluate data analytics algorithms using information retrieval metrics. As an important case study for such synthetic log repository, we populate an event-log repository with complex user-behavior instances in the healthcare domain, where the behavior is defined as a sequence of events by a user. Since user behavior has a complex nature, we defined a user-behavior pattern language (BPL) that allows the domain experts to represent both the desired behavior patterns that is used by the log generator engine, and for defining a target user-behavior pattern to be searched in the generated log repository. We use constraint-based approximate event-pattern matching techniques to search and identify the instances of the target pattern in the repository. In this paper, we introduce our BPL, present our event-log generator engine and the produced log repository, and use our pattern matching algorithm to identify the extracted user behaviours in the repository to show the practicality and usefulness of our proposed framework.