Evidence-based managerial decision-making with machine learning: The case of Bayesian inference in aviation incidents

Understanding the factors behind aviation incidents is essential, not only because of the lethality of the accidents but also the incidents’ direct and indirect economic impact. Even minor incidents trigger significant economic damage and create disruptions to aviation operations. It is crucial to investigate these incidents to understand the underlying reasons and hence, reduce the risk associated with physical and financial safety in a precarious industry like aviation. The findings may provide decision-makers with a causally accurate means of investigating the topic while untangling the difficulties concerning the statistical associations and causal effects. This research aims to identify the significant variables and their probabilistic dependencies/relationships determining the degree of aircraft damage. The value and the contribution of this study include (1) developing a fully automatic ML prediction based DSS for aircraft damage severity, (2) conducting a deep network analysis of affinity between predicting variables using probabilistic graphical modeling (PGM), and (3) implementing a user-friendly dashboard to interpret the business insight coming from the design and development of the Bayesian Belief Network (BBN). By leveraging a large, real-world dataset, the proposed methodology captures the probability-based interrelations among air terminal, flight, flight crew, and air-vehicle-related characteristics as explanatory variables, thereby revealing the underlying, complex interactions in accident severity. This research contributes significantly to the current body of knowledge by defining and proving a methodology for automatically categorizing aircraft damage severity based on flight, aircraft, and PIC (pilot in command) information. Moreover, the study combines the findings of the Bayesian Belief Networks with decades of aviation expertise of the subject matter expert, drawing and explaining the association map to find the root causes of the problems and accident relayed variables.