Vestibular autorotation test in primary selection of military student pilots using neural networks.

BACKGROUND: Selection of student pilots to the Royal Danish Air Force involves a physical examination including a vestibular test. Usually tests for selection proposes are not well documented. HYPOTHESIS: The result of the vestibular autorotation test (VAT) is correlated to the ability to learn to fly in a military context. METHODS: A Multi Layer Perception neural network with three layers configured as a Back Propagation Network was tested using data originating from horizontal VAT of 59 student pilot candidates, given the outcome of the pre-jet basic flight check. In the analysis the leave-one-out method was used. RESULTS: Based on horizontal data only the network correctly classified the student pilot candidates as having been passed or rejected within a verification error margin < 0.1. CONCLUSION: The result indicates that VAT performed at the initial physical examination is a powerful tool for the elimination of unfit student pilot candidates when data are analyzed in neural networks.