Prediction of Injuries and Fatalities in Aviation Accidents through Machine Learning

Abstract

This paper concerns application of various machine learning techniques to derive classification models for predicting conditions that increase the likelihood of aviation accidents involving fatalities and serious injuries. Machine learning classification techniques, including Decision Trees, K-Nearest Neighbors, Support Vector Machines (SVMs), and Artificial Neural Networks (ANNs) are applied to datasets derived from original data obtained from Federal Aviation Administration (FAA) Aviation Accident and Incident Records from 1975-2002. The accident data are filtered to focus on FAA Part 91 (General Aviation) accidents involving powered, fixed-wing, manufactured aircraft. The results demonstrate ANNs to yield the most accurate prediction levels for both fatal accidents and accidents involving severe injuries. The results also demonstrate that machine learning approaches may yield insightful information beyond what is available through traditional statistical analysis methodologies.