To address the limitations of Next Generation Radar-based bird strike forecasting, this study modeled 12 spatiotemporal weather features from the National Oceanic and Atmospheric Administration alongside bird strike risk using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), XGBoost regression tree, and Bayesian network algorithms. Five years of bird strike data from four geographically diverse airfields served as the target risk variable, categorized as low, moderate, or severe based on Department of the Air Force risk models. The ensemble model, which combines the LSTM-RNN and XGBoost regression algorithms, yielded the most accurate forecasts, achieving 80% to 93% accuracy across all airfields, with F1 scores of approximately 80%. Individual model accuracy ranged from 59% to 93%. Confidence intervals from the Categorical Gaussian Bayesian Network (CGBN) were the most stable due to entropy-based estimation, outperforming the bootstrapping approach used in LSTM-RNN and XGBoost. Receiver Operating Characteristic (ROC) curves showed consistent patterns across oversampling techniques and locations, suggesting that imputation and sampling procedures introduced minimal bias. The three oversampling techniques utilized to balance the data did not significantly affect model outcomes.
The study employed the Analytic Hierarchy Process (AHP) to incorporate human expertise into the modeling process by using priority rankings from the United States Department of Agriculture Airport Wildlife Hazard Biologists. The AHP-derived feature list was statistically distinct from the machine learning rankings, indicating a divergence between expert preferences for long-term environmental indicators and the models’ emphasis on short-term weather patterns with immediate predictive power.
These findings validate the effectiveness of machine learning for bird strike forecasting and demonstrate a replicable, data-driven framework for aviation risk assessment. By enabling accurate, location-specific risk predictions hours in advance, the models offer a proactive approach to augment current radar-based systems, which are currently limited to near-real-time detection. This approach supports risk-informed decision-making for mission planning, airfield operations, and in-flight route adjustments, particularly in high-threat environments. Furthermore, the integration of environmental data, expert input, and machine learning yields a tactical feature set that enhances hazard mitigation strategies and advances the application of predictive analytics in aviation safety. This model contributes to the evolution of adaptive airspace management and risk-based resource allocation, enhancing the resiliency of both military and civilian flight operations. The complete Python script can be found at the GitHub location (https://doi.org/10.5281/zenodo.15760325).
","abstract_html":"<p>To address the limitations of Next Generation Radar-based bird strike forecasting, this study modeled 12 spatiotemporal weather features from the National Oceanic and Atmospheric Administration alongside bird strike risk using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), XGBoost regression tree, and Bayesian network algorithms. Five years of bird strike data from four geographically diverse airfields served as the target risk variable, categorized as low, moderate, or severe based on Department of the Air Force risk models. The ensemble model, which combines the LSTM-RNN and XGBoost regression algorithms, yielded the most accurate forecasts, achieving 80% to 93% accuracy across all airfields, with F1 scores of approximately 80%. Individual model accuracy ranged from 59% to 93%. Confidence intervals from the Categorical Gaussian Bayesian Network (CGBN) were the most stable due to entropy-based estimation, outperforming the bootstrapping approach used in LSTM-RNN and XGBoost. Receiver Operating Characteristic (ROC) curves showed consistent patterns across oversampling techniques and locations, suggesting that imputation and sampling procedures introduced minimal bias. The three oversampling techniques utilized to balance the data did not significantly affect model outcomes.</p> <p>The study employed the Analytic Hierarchy Process (AHP) to incorporate human expertise into the modeling process by using priority rankings from the United States Department of Agriculture Airport Wildlife Hazard Biologists. The AHP-derived feature list was statistically distinct from the machine learning rankings, indicating a divergence between expert preferences for long-term environmental indicators and the models’ emphasis on short-term weather patterns with immediate predictive power.</p> <p>These findings validate the effectiveness of machine learning for bird strike forecasting and demonstrate a replicable, data-driven framework for aviation risk assessment. By enabling accurate, location-specific risk predictions hours in advance, the models offer a proactive approach to augment current radar-based systems, which are currently limited to near-real-time detection. This approach supports risk-informed decision-making for mission planning, airfield operations, and in-flight route adjustments, particularly in high-threat environments. Furthermore, the integration of environmental data, expert input, and machine learning yields a tactical feature set that enhances hazard mitigation strategies and advances the application of predictive analytics in aviation safety. This model contributes to the evolution of adaptive airspace management and risk-based resource allocation, enhancing the resiliency of both military and civilian flight operations. The complete Python script can be found at the GitHub location (https://doi.org/10.5281/zenodo.15760325).</p>","abstract_has_math":false,"creators":["Powell, Jason Anthony"],"institution":null,"degree_name":"Doctor of Philosophy in Aviation","degree_level":"Dissertation - Open Access","degree_discipline":"College of Aviation","degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-07-01T07:00:00Z","date_published":"2025-07-01T07:00:00Z","updated_at":"2026-07-27T19:26:16Z","subjects":["aircraft bird strike risk","aircraft safety","analytical hierarchy process","Bayesian network","deep learning","LSTM-RNN","weather prediction","Artificial Intelligence and Robotics","Aviation Safety and Security"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://commons.erau.edu/edt/923","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Powell, Jason Anthony"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"thesis:degree_discipline","label":"Discipline","values":["College of Aviation"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation - Open Access"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy in Aviation"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["aircraft bird strike risk","aircraft safety","analytical hierarchy process","Bayesian network","deep learning","LSTM-RNN","weather prediction","Artificial Intelligence and Robotics","Aviation Safety and Security"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://commons.erau.edu/edt/923"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["To address the limitations of Next Generation Radar-based bird strike forecasting, this study modeled 12 spatiotemporal weather features from the National Oceanic and Atmospheric Administration alongside bird strike risk using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), XGBoost regression tree, and Bayesian network algorithms. Five years of bird strike data from four geographically diverse airfields served as the target risk variable, categorized as low, moderate, or severe based on Department of the Air Force risk models. The ensemble model, which combines the LSTM-RNN and XGBoost regression algorithms, yielded the most accurate forecasts, achieving 80% to 93% accuracy across all airfields, with F1 scores of approximately 80%. Individual model accuracy ranged from 59% to 93%. Confidence intervals from the Categorical Gaussian Bayesian Network (CGBN) were the most stable due to entropy-based estimation, outperforming the bootstrapping approach used in LSTM-RNN and XGBoost. Receiver Operating Characteristic (ROC) curves showed consistent patterns across oversampling techniques and locations, suggesting that imputation and sampling procedures introduced minimal bias. The three oversampling techniques utilized to balance the data did not significantly affect model outcomes.
The study employed the Analytic Hierarchy Process (AHP) to incorporate human expertise into the modeling process by using priority rankings from the United States Department of Agriculture Airport Wildlife Hazard Biologists. The AHP-derived feature list was statistically distinct from the machine learning rankings, indicating a divergence between expert preferences for long-term environmental indicators and the models’ emphasis on short-term weather patterns with immediate predictive power.
These findings validate the effectiveness of machine learning for bird strike forecasting and demonstrate a replicable, data-driven framework for aviation risk assessment. By enabling accurate, location-specific risk predictions hours in advance, the models offer a proactive approach to augment current radar-based systems, which are currently limited to near-real-time detection. This approach supports risk-informed decision-making for mission planning, airfield operations, and in-flight route adjustments, particularly in high-threat environments. Furthermore, the integration of environmental data, expert input, and machine learning yields a tactical feature set that enhances hazard mitigation strategies and advances the application of predictive analytics in aviation safety. This model contributes to the evolution of adaptive airspace management and risk-based resource allocation, enhancing the resiliency of both military and civilian flight operations. The complete Python script can be found at the GitHub location (https://doi.org/10.5281/zenodo.15760325).
"]},{"key":"dc:title","label":"Title","values":["Aircraft Bird Strike Risk Prediction Using Machine Learning and Analytic Hierarchy Process"]}]}],"canonical_facts":{"dc:creator":["Powell, Jason Anthony"],"dc:description.abstract":["To address the limitations of Next Generation Radar-based bird strike forecasting, this study modeled 12 spatiotemporal weather features from the National Oceanic and Atmospheric Administration alongside bird strike risk using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), XGBoost regression tree, and Bayesian network algorithms. Five years of bird strike data from four geographically diverse airfields served as the target risk variable, categorized as low, moderate, or severe based on Department of the Air Force risk models. The ensemble model, which combines the LSTM-RNN and XGBoost regression algorithms, yielded the most accurate forecasts, achieving 80% to 93% accuracy across all airfields, with F1 scores of approximately 80%. Individual model accuracy ranged from 59% to 93%. Confidence intervals from the Categorical Gaussian Bayesian Network (CGBN) were the most stable due to entropy-based estimation, outperforming the bootstrapping approach used in LSTM-RNN and XGBoost. Receiver Operating Characteristic (ROC) curves showed consistent patterns across oversampling techniques and locations, suggesting that imputation and sampling procedures introduced minimal bias. The three oversampling techniques utilized to balance the data did not significantly affect model outcomes.
The study employed the Analytic Hierarchy Process (AHP) to incorporate human expertise into the modeling process by using priority rankings from the United States Department of Agriculture Airport Wildlife Hazard Biologists. The AHP-derived feature list was statistically distinct from the machine learning rankings, indicating a divergence between expert preferences for long-term environmental indicators and the models’ emphasis on short-term weather patterns with immediate predictive power.
These findings validate the effectiveness of machine learning for bird strike forecasting and demonstrate a replicable, data-driven framework for aviation risk assessment. By enabling accurate, location-specific risk predictions hours in advance, the models offer a proactive approach to augment current radar-based systems, which are currently limited to near-real-time detection. This approach supports risk-informed decision-making for mission planning, airfield operations, and in-flight route adjustments, particularly in high-threat environments. Furthermore, the integration of environmental data, expert input, and machine learning yields a tactical feature set that enhances hazard mitigation strategies and advances the application of predictive analytics in aviation safety. This model contributes to the evolution of adaptive airspace management and risk-based resource allocation, enhancing the resiliency of both military and civilian flight operations. The complete Python script can be found at the GitHub location (https://doi.org/10.5281/zenodo.15760325).
"],"dc:identifier":["https://commons.erau.edu/edt/923"],"dc:subject":["aircraft bird strike risk","aircraft safety","analytical hierarchy process","Bayesian network","deep learning","LSTM-RNN","weather prediction","Artificial Intelligence and Robotics","Aviation Safety and Security"],"dc:title":["Aircraft Bird Strike Risk Prediction Using Machine Learning and Analytic Hierarchy Process"],"thesis:degree_discipline":["College of Aviation"],"thesis:degree_level":["Dissertation - Open Access"],"thesis:degree_name":["Doctor of Philosophy in Aviation"]},"updated_at":"2026-07-27T19:26:16Z"}