Abstract:

Viewing the process model in event logs as the combination of the two adjacent activities, two novel process models are proposed. First, the process model is extracted by constructing adjacency matrix, taking advantage of the adjacency relationships of activities in the event logs. To improve the prediction accuracy of the model, loops are avoided in the process model. So, each activity in this model will only happen once. Second, the serial number of activities in the event logs to the adjacency matrix is added, constructing a new adjacency matrix with sequence information. Based on the new adjacency matrix, the process model is extracted. Each activity in this model will only happen once at the same sequence position. Third, with the adjacency matrix, the path from each prior node to next nodes in the process model and their corresponding probabilities are gotten. Then, according to the type and characteristic information of the event logs, predictions of the information are made which are needed by decision makers, e.g. activity name, waiting time, and probability based on process model. Finally, the effectiveness of the proposed models by comparing the prediction results of random data and real data based on process models is verified.