Is this project an undergraduate, graduate, or faculty project?
Graduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Bill Deng Pan, Graduate student Yupeng Yang, Graduate student
Lead Presenter's Name
Bill Deng Pan
Lead Presenter's College
DB College of Aviation
Faculty Mentor Name
Dr. Dahai Liu
Abstract
Runway incursions remain one of the most persistent safety challenges in modern aviation operations, often resulting from complex interactions among human, environmental, and operational factors. While existing Safety Management System (SMS) frameworks emphasize monitoring the frequency of incursions, they provide limited predictive capability regarding event severity. This ongoing study seeks to address that gap by applying a novel tabular-to-image deep-learning approach to improve the accuracy and interpretability of runway-incursion severity prediction models. Data for this study will be drawn from the Federal Aviation Administration (FAA) Runway Safety Statistics and National Transportation Safety Board (NTSB) accident databases. The methodology will involve transforming structured aviation data into two-dimensional image representations based on normalized feature correlations among operational and environmental variables such as visibility, wind, aircraft type, airport class, and time of day. This conversion will allow convolutional neural networks (CNNs) to capture intricate, nonlinear relationships between predictors that are often missed by conventional tabular models. Multiple CNN architectures will be trained and compared to traditional machine-learning classifiers using cross-validation and key performance metrics, including accuracy, precision, recall, and F1 score. Model interpretability will be enhanced through feature importance visualization techniques to identify the most influential variables contributing to severity classification. This research is expected to demonstrate that tabular-to-image learning can significantly improves classification accuracy and sensitivity across different severity levels of runway incursions. The findings of this study aim to strengthen predictive safety analytics by advancing from reactive reporting to proactive risk management. The proposed framework has the potential to enhance SMS capabilities by providing airport operators, air traffic controllers, and regulatory agencies with data-driven tools for anticipating and mitigating high-severity runway-incursion events, contributing to a safer and more resilient airport operations.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
Advancing Runway Incursion Severity Prediction through Tabular-to-Image Deep Learning
Runway incursions remain one of the most persistent safety challenges in modern aviation operations, often resulting from complex interactions among human, environmental, and operational factors. While existing Safety Management System (SMS) frameworks emphasize monitoring the frequency of incursions, they provide limited predictive capability regarding event severity. This ongoing study seeks to address that gap by applying a novel tabular-to-image deep-learning approach to improve the accuracy and interpretability of runway-incursion severity prediction models. Data for this study will be drawn from the Federal Aviation Administration (FAA) Runway Safety Statistics and National Transportation Safety Board (NTSB) accident databases. The methodology will involve transforming structured aviation data into two-dimensional image representations based on normalized feature correlations among operational and environmental variables such as visibility, wind, aircraft type, airport class, and time of day. This conversion will allow convolutional neural networks (CNNs) to capture intricate, nonlinear relationships between predictors that are often missed by conventional tabular models. Multiple CNN architectures will be trained and compared to traditional machine-learning classifiers using cross-validation and key performance metrics, including accuracy, precision, recall, and F1 score. Model interpretability will be enhanced through feature importance visualization techniques to identify the most influential variables contributing to severity classification. This research is expected to demonstrate that tabular-to-image learning can significantly improves classification accuracy and sensitivity across different severity levels of runway incursions. The findings of this study aim to strengthen predictive safety analytics by advancing from reactive reporting to proactive risk management. The proposed framework has the potential to enhance SMS capabilities by providing airport operators, air traffic controllers, and regulatory agencies with data-driven tools for anticipating and mitigating high-severity runway-incursion events, contributing to a safer and more resilient airport operations.