Research On Conformance Analysis Methods Of Process Mining

Many companies have adopted Process Management System (PMS) to support their business process in different forms. On the one hand, these systems log events during process executions. There exists a predefined process model to describe how the business process should be carried out in practice. This leads to a question of whether the logs in these process management systems are consistent with those of the predefined process models. Conformance analysis is also called conformance checking, which aims to evaluate the conformance between the process model and the log associated with the model, and to measure the degree of conformance between them, and find out where inconsistencies exists for optimizing the model.In this paper, we focus on the control flow of the model, and analyze the conformance between the model and the log from different perspectives. Three metrics are used to measure the conformance, that is, fitness, behavioral appropriateness and structural appropriateness. Fitness is used to check whether the traces in the log are consistent with the control flow in the model, that is, whether the traces in the log can "pass" the model. Behavioral appropriateness is used to evaluate the accuracy when the model describes behaviors in the log that is, whether the behaviors allowed by the model are actually executed and recorded in the log. Structural appropriateness is used to evaluate the complexity of the model itself, whether the model is simple and clear.In this paper, we propose different algorithms to quantify these three metrics. And for the same metric, we analyze the shortcomings of the current popular algorithms, improve them, and then use the improved algorithms to design the conformance checking plug-in, and do experiments to verify these algorithms in the ProM framework. The main improvements are as follows:Fitness:new algorithm simplifies statistical data and takes into account the effect of the fitness of a single path in a log for the overall fitness of the log.Behavioral appropriateness:new algorithm deals with "long cycle" structure of the model better, and reduces the difficulty of the model structure analysis.Structural appropriateness:new algorithm takes into account the effect of implicit places of a model for this model’s structural appropriateness.Based on the improved algorithms, this paper carries out the integration of the three conformance metrics, and presents a comprehensive metric. In this metric, this paper fully considers the basic function of fitness in conformance checking. Only under the premise that the model’s fitness metric is 1, will we continue to evaluate behavioral appropriateness and structural appropriateness. And through the comprehensive metric, we chose the best model for each log used in the paper.Finally, we summarize the work done in this paper, and point out the deficiencies of the work. Then we expound the difficulties in the field of conformance checking, and make forecasts about the future work.