Log Clustering-based Method for Repairing Missing Traces with Context Probability Information

Abstract

In real business processes, low quality event logs due to outliers and missing values tend to degrade the performance of process mining related algorithms, which in turn affects the correct execution of decisions. In order to repair the missing values in event logs under the condition that the reference model of the process system is unknown, this paper proposes a method that can repair consecutive missing values. First, the event logs are divided according to the integrity of the trace, and then the cluster algorithm is applied to complete logs to generate homogeneous trace clusters. Then match the missing trace to the most similar sub log, generate candidate sequences according to the context of the missing part, calculate the context probability of each candidate sequence, and select the one with the highest probability as the repair result. When the number of missing items in the trace is 1, our method has the highest repair accuracy of 97.5 percent in the Small log and 93.3 percent in the real event logs bpic20. Finally, the feasibility of this method is verified on four event logs with different missing ratios and has certain advantages compared with existing methods.