Research On Conformance Algorithm For Process Mining

As the business process becomes more and more complex,the log data recorded in the business process is growing at an amazing speed.How to use these log information to improve the business process is becoming more and more important for the enterprise.Process mining mainly analyzes the event logs of business processes,and discovers,detects,and improves system business processes based on real event logs generated by business processes.The conformance is used to check whether there is any deviation between the process model and the event log.The main metrics are the degree of fitness,precision,generalization,and simplicity.The precision of the process model is the degree to which the model accurately expresses the behavior of the log.Existing precision algorithms generally have problems with inaccurate calculation results.The generalization is a metric to prevent overfitting of the process model in the process mining.Because the computational of generalization needs to consider unknown behaviors,there are still many problems with the existing generalization algorithms.An LMA algorithm that combines the log automaton and the model automaton is proposed to measure the precision,for the shortcomings of the precision automaton algorithm that can not fully explore the deviation states.The maximum length of the full execution sequences of the Petri net model is limited,according to the fact that the loop structure of the model could not be executed without any limit.The log automaton is constructed from the event log that is fully fit after alignment.Then,the model automaton is constructed based on the log automaton and the full transition execution sequences of the Petri net model.An escaping state is a state that appears in the model automaton and does not appear in the log automaton.The precision can be determined by the proportion of the escaping states.The LMA algorithm does not replay the process model based on the event log,and it can explore more deviations than the precision automaton algorithm,and the obtained precision value is more accurate.A generalization algorithm based on generalization automaton is proposed,for the existing generalization algorithm that based on alignment has shortcomings of relies on Bayesian distribution and high time complexity.According to the idea of the process tree algorithm that the generalization is computed from the frequencies of nodes visited in the tree,and the idea of the generalization algorithm that based on alignment calculates the generalization based on the states.The fully fit event log is replayed on the process model.A generalization automaton is constructed according to the change of the marking state.The number of times the states is visited and the firing set of activities are recorded.The higher the ratio of the number of the state is visited and the number of activities that the state fires,the more reliable of the state,and the smaller the likelihood of firing new activity when the state is visited the next time,and the higher the generalization.The generalization automaton algorithm does not depend on any probability distribution,and does not handle each event individually,and the time complexity is lower than other algorithms.According to the two algorithms proposed in this paper,simulation experiments of the same model and the different structural models are designed to calculate the precision,generalization,and the time spent by the algorithm.The comparison with other classical algorithms is given,and the correctness and practicality of the algorithms is illustrated.