Probabilistic prediction model of air traffic controllers' sequencing strategy based on pairwise comparisons

Abstract

Sequencing arrival flights is a major task of air traffic management, and there exist various optimization tools to support the air traffic controllers. It is, however, difficult to employ these tools in the actual operational environments since they lack consideration on the human cognitive process. This paper proposes a new framework to predict the arrival sequences based on a preference learning approach, where we learn the sequence data operated by human controllers. The proposed algorithm works in two-stages: it first learns the pairwise preference functions between arrivals using binomial logistic regression, and then it induces the total sequence for a new set of arrivals by comparing the scores of each aircraft, which are the sums of pairwise preference probabilities. The proposed model is demonstrated with real traffic data at Incheon International Airport and its performance is assessed using the Spearman's rank correlation.