Abstract: The problem of predicting an ongoing business process based on sequence is discussed. Business process predictive monitoring is an important part of process mining. Most of the existing research focuses on forecasting based on explicit workflow models. However, the enterprise may not implement end-to-end workflow modeling and management for the process, or only own partial execution logs due to permissions. In these cases, it is difficult to make prediction based on the complete process model. This article proposes a frequent activity set based sequence encoding method to handle the low-frequency activities in the log, and performs prediction by searching historical similar data. As the log is updated, the algorithm will adapt to model changes due to the concept drift. The algorithm is validated on real data sets.

Key words: process mining, concept drift, sequence coding, predictive monitoring