Advanced Statistical Learning and Prediction of Complex Runway Incursion

In 2015, 1,507 runway incursions capable of inducing collisions occurred at airports in the United States, so it is obviously very important to identify significant factors underlying such incursions, to predict potential runway incursion occurrences, and to prepare systematic programs for reducing the number of incursions and prevent runway collisions. Presence of a large volume of data, multiple variables, and complex interactions among them pose a significant challenge to resolving this problem. To tackle this challenge, we developed a data-driven prediction model using a component of advanced statistical theory, i.e., a generalized additive model (GAM). GAM can account for flexible modeling of multiple variables over a broad range of modeling distributions. We obtained, parsed, and transformed various predictor variables from many heterogeneous databases to create interpretable datasets for statistical modeling. We demonstrated promising performance of GAM while making systematic investigations into prediction accuracy of runway incursion at United States airports (including all types of commercial, military, and other general data). Results show that GAM can identify critical factors (airport complexity, number of operations, and visibility) in predicting a number of the runway incursions. Performance comparison of two popular GAM smoothers (i.e., cubic regression splines and thin plate regression splines) has demonstrated promising accuracy of both methods. These results imply that statistical predictions developed using GAM will help in better prediction of runway incursion when more data become available in the future.