Research on safety assessment of air traffic control in small and medium airports based on machine learning

Abstract

To establish an impartial air safety evaluation system, this study translated qualitative air safety assessment into quantitative probability estimation using machine learning and historical data. A quantitative ATC safety assessment framework was formulated based on the SHEL model, complemented by a cloud model for safety evaluation drawing on fuzzy and uncertainty theories. A copula function analyzed correlations among cloud model indices, refined the model, and the entropy weight method determined membership weights. Ordered logistic regression categorized ATC safety levels, while genetic algorithms extracted factors' attributes and principal component analysis reduced model complexity. Ultimately, a semi-supervised learning-based collaborative ATC safety evaluation system was developed, enhancing the cloud model's generalizability and precision. Cross-validation and multifaceted verification confirmed the system's objectivity and reliability.