A Predictive Ensemble Model to Minimize Bird Strike Occurrences on Aircrafts at U.S. Airports
Open AccessBird strikes against inbound or outgoing aircraft at airports around the world have been a well-documented phenomenon directly impacting flight safety for decades and continues to be a notable problem to this day. However, there are limited attempts by prior researchers to address this problem by applying Advanced Machine Learning techniques to accurately predict the probability of an aircraft strike given certain flight operation conditions. This Praxis addresses this issue by applying Random Forest, Gradient Boosting and Ensemble Machine Learning modeling against the bird strike problem. In this Praxis, Ensemble models outperformed Random Forest by at least 5% when applied to bird strike data at Denver International Airport (DIA) using the following model comparison and validation techniques: Area under the ROC Curve (AUC-ROC), F1 Score, Precision, Recall, Matthew’s Correlation Coefficient & Log Loss. Ensemble models also consistently outperformed Gradient Boosting models when applied to bird strike data at DIA, but not by the 7% that was originally hypothesized. Moreover, for the models created, an aircraft’s physical dimensions as well as date and time were primary factors that directly correlated to bird strikes at DIA given this particular merged dataset, and the application of the Synthetic Minority Over-sampling Technique (SMOTE) was necessary to address the highly imbalanced dataset involved with this project. Ultimately, recommendations for future research include replicating identical Machine Learning model approaches at different airports so future research could compare and tease out other key factors directly applicable to accurately predicting bird strikes on aircraft. Second, future researchers could create a more effective tool to consolidate current flight data with critical bird strike information. Right now, there is no holistic database that allows for effective, efficient, and accurate consolidation of both positive bird strikes and negative bird strikes into one repository. Finally, future researchers could leverage other imbalanced data analytic techniques such as under-sampling or applying a combination over-sampling the minority samples and under-sampling the majority samples.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.