Learning Bayesian Network from Event Logs Using Mutual Information Test

A Bayesian network can be considered to be a powerful tool for various analyses (e.g. inference analysis, sensitivity analysis, evidence propagation, etc.), however, it is first necessary to obtain the Bayesian network structure of a given dataset, and this, an NP hard problem, is not an easy task. Among the available scoring metrics, the present study employed Mutual Information Test (MIT) to construct a Bayesian network from the event logs of port logistics data covering six days of observations. Additionally, dynamic programming was used to shorten the combinatorial calculation of the metrics and, later, to minimize the computation time. To validate our method, we conducted a case study of port processes using actual event logs from an Asian port.