Authors: Shlok Misra; Ila Toppo; Flavio Antonio Coimbra Mendonca
Addresses: Embry-Riddle Aeronautical University, 1 Aerospace Blvd., Daytona Beach, FL, 32114, USA ' Embry-Riddle Aeronautical University, 1 Aerospace Blvd., Daytona Beach, FL, 32114, USA ' Embry-Riddle Aeronautical University, 1 Aerospace Blvd., Daytona Beach, FL, 32114, USA
Abstract: The objectives of this study were to identify the factors that are statistically associated with the probability of aircraft damage in the event of a bird strike and to develop classification models to predict aircraft damage in an event of a bird strike. The FAA National Wildlife Strike Database was used for the study to develop random forest, artificial neural network, logistic regression, support vector machine, extra gradient boost (XGBoost), and K-nearest neighbours classifier models. The random forest classifier, logistic regression, and XGBoost classifier exhibited the most robust predictive powers with accuracies of 78.81%, 78.51% and 78.35%, respectively. Based on the variable assessment scores for the random forest classifier, the size of the bird, height of impact, aircraft speed, and aircraft mass had the highest contributions towards predicting aircraft damage for the model.
Keywords: machine learning in aviation; bird strikes; wildlife strikes; risk assessment; wildlife hazard management; aviation safety; classification models; FAA National Wildlife Strike Database; data mining.
International Journal of Sustainable Aviation, 2022 Vol.8 No.2, pp.136 - 151
Received: 01 Sep 2021
Accepted: 05 Nov 2021
Published online: 19 Apr 2022 *