Abstract:
Human factors are increasingly leading the General Aviation’s (GA) accident
causation although the total number of accidents have significantly improved over past few
decades. The actions majorly taken so far correspond to reactive safety approaches rather
than proactive ones. GA has been neglected a lot in terms of safety and risk mitigation as
the fatality rate has been almost constant for many years now. In this research study, the
probable causes of GA Loss of Control-In Flight (LOC-I) accidents under Initial Climb
(ICL) phase of flight are obtained from National Transportation Safety Board (NTSB).
Each accident is classified into one of the 9 LOC-I accident categories defined by
International Air Transport Association (IATA). The preprocessed and feature engineered
dataset is fed to a Random Forest (RF) model to be trained. The prediction model gives an
accuracy and F-1 score of 88% on the test set. Feature importance and SHapley Additive
exPlanation (SHAP) analysis of RF model is performed to get the most influencing features
on prediction. The most influential features of the RF model vulnerable are connected to
the Human Factor Analysis and Classification System (HFACS) to get insights into the
most vulnerable HFACS levels.
