Application of Bayesian networks and information theory to estimate the occurrence of mid-air collisions based on accident precursors