An Event-Level Clustering Framework for Process Mining Using Common Sequential Rules

Abstract

Process mining techniques extract useful knowledge from event logs to analyse and improve the quality of process execution. However, size and complexity of the real-world event logs make it difficult to apply standard process mining techniques, thus process discovery results in spaghetti-like models which are difficult to analyse. Several event abstraction techniques are developed to group-up low-level activities into higher level activities, but abstraction ignores the low level critical process details in the real-world business scenarios. Also, trace clustering techniques have been extensively used in literature to cluster the processes executions which are homogeneous in nature, but event-level clustering is not yet considered for process mining. In this paper, a novel framework is proposed to identify event-level clusters in a business process log by decomposing into several sub-logs based upon the similarity of the sequences between events. Our technique provides clustering without abstraction of very large complex event logs. Proposed algorithm Common Events Identifier (CEI) is applied on a real-world telecommunication log and the results are compared with two well-known trace clustering techniques from the literature. Our results achieved high accuracy of clustering and improved the quality of resulting process models using the given size and complexity of the event log. We further demonstrated that the proposed techniques improved process discovery and conformance results for a given event log.Keywords: process mining • process discovery • trace clustering • rulebased mining • process analytics • business processes • sequence mining.