Alignment Based Precision Checking

Abstract

Abstract. Most organizations have process models describing how cases need to be handled. In fact, legislation and standardization (cf. the Sarbanes-Oxley Act, the Basel II Accord, and the ISO 9000 family of standards) are forcing organizations to document their processes. These processes are often not enforced by information systems. However, torrents of event data are recorded by today's information systems. These recorded events reflect how processes are really executed. Often reality deviates from the modeled behavior. Therefore, measuring the extent process executions conform to a predefined process model is increasingly important. In this paper, we propose an approach to measure the precision of a process model with respect to an event log. Unlike earlier approaches, we first align model and log thus making our approach more robust, even in case of deviations. The approach has been implemented in the ProM 6 tool and evaluated using both artificial and real life cases.

Measuring precision of modeled behavior

“…The technique considers 'one' or 'all' optimal alignments, and is based on the methods described in [24][25][26]. However, there is a fundamental difference: whereas in [24][25][26] precision is measured based on log-based model replay, the approach in this section is based on alignments [9]. The advantages are manifold.…”

“…Most of the existing approaches, e.g., [9,24], construct an automaton from the beginning of the traces forward. Thus, choices that are made in the beginning of traces may have bigger influence on precision value than choices that are made towards the end of traces.…”

“…Then we use these alignments to measure precision between the original log and the model. In this paper, we generalize the approach presented in [9] by introducing various possible ways of computing precisions based on alignments, their log completeness requirements, and their issues in order to obtain accurate precision values.…”

Event-Based Real-Time Decomposed Conformance Analysis

“…One of the drawbacks of this approach is that for undeterministic models, the heuristics used in the replay may lead to overestimating the metrics, due to the arti cial creation of super uos tokens in the model. To overcome this fundamental problem, Adriansyah et al propose an alternative approach where the concept of alignments are introduced in order to match an event trace with a path through the model as closely as possible [2,3,4].…”

Improving Business Process Models Using Observed Behavior

“…Precision compares the state space of the tree execution while replaying the log. Our metric is inspired by [5] and counts so-called escaping edges, i.e. decisions that are possible in the model, but never made in the log.…”