Flight Safety Research Based on GBDT Regression and Graph Neural Networks

This paper focuses on an in-depth study of flight safety in the civil aviation transportation industry. Considering the great impact of serious flight accidents on the safety of airlines and passengers, this study first reviewed the public information of the Civil Aviation Administration of China (CAAC) and conducted a reliability study on the relevant data. Firstly, we questioned the problems in the data set through mathematical and statistical methods and found that the “elevation and pitch rate” is a key factor affecting flight safety using GBDT regression analysis. Secondly, using a graphical neural network model enhanced by cluster analysis and attention mechanism, this paper identifies the main operational factors affecting aircraft flight as stick volume and attitude, and points out that incorrect stick release operation is the cause of abnormal landing G-value. Finally, this paper explores flight overrun events through qualitative and quantitative analysis and finds that most overrun events occur at landing, especially on the takeoff-to-landing route from Airport 68 to Airport 9. Although the chi-square test showed that aircraft number and route had a significant effect on the type of overrun, the effect was small. These findings provide an important basis for understanding and improving flight safety issues.