Accepted paper at ICPM 2020 Workshop SA4PM 2020

Authors

Florian Richter, Andrea Maldonado, Ludwig Zellner, Thomas Seidl

1st International Workshop on Streaming Analytics for Process Mining (SA4PM 2020)
in conjunction with the 2nd International Conference on Process Mining (ICPM 2020),
04–09 October 2020, Virtual

Abstract

Identifying structures in data is an essential step to enhance insights and understand applications. Clusters and anomalies are the basic building blocks for those structures and occur in various types. Clusters vary in shape and density, while anomalies occur as single-point outliers, contextual or collective anomalies. In online applications, clusters even have a higher complexity. Besides static clusters, which represent a persistent structure throughout the whole data stream, many clusters are dynamic, tend to drift and are only observable in certain time frames. Here, we propose OTOSO, a monitoring tool based on OPTICS. OTOSO is an anytime structure visualizer, that plots representations for density-based trace clusters in process event streams. It identifies temporal deviation clusters and visualizes them as a time-dependent graph. Each node represents a cluster of traces by size and density. Edges yield information about merging and splitting trace clusters. The aim is to provide an on-demand overview over the temporal deviation structure during the process execution. Not only for online applications but also for static datasets, our approach yields insights about temporally limited occurrences of trace clusters, which are difficult to detect using a global clustering approach.