Understanding General Aviation Accidents in Terms of Safety Systems

This research provides a new, data-driven method that could be used to estimate the impact of integrating an automated safety system into a target class of general aviation aircraft. General Aviation (GA), that is, air travel apart from scheduled air carriers, is still more dangerous than automobile travel by several metrics. This fact should drive research to understand which safety systems might make significant improvements in GA aircraft safety. Pre-existing accident classification schemes are often very general and do not necessarily provide the insight necessary to judge the impact of a given safety technology. This paper attempts to use machine learning methods, applied to a novel transformation of the publicly available NTSB accident database, to create a model based on a set of pre-scored accident records that can be used to provide a notional estimate of the impact of an automatic ground collision avoidance system (Auto GCAS) in terms of fatal events that might have been prevented had such a system been installed. This study found that the number of fatality accidents that were predicted to be prevented by Auto GCAS was significant. The events that were thus predicted by the model spanned multiple CICTT Occurrence Categories, indicating that attempting to categorize the impact of the Auto GCAS system in terms of controlled flight into terrain (CFIT) alone, for example, would under-represent the potential benefit by not including saves cutting across the low altitude operations (LALT), unintended flight into instrument meteorological conditions (UIMC), and loss of control in-flight (LOC-I) categories.