Abstract
The construction of an accurate aviation safety prediction model to determine the change pattern of accidents and their causal factors is of great significance for intelligent management and proactive decision-making in aviation safety. To this end, a random forest algorithm based on a combination of Bow-tie models is proposed in this paper for aviation safety causal prediction, which completes the optimization of safety prediction model parameters and the ranking of causal variable contributions. Firstly, the Bow-tie model is introduced to determine correlation identification of aviation safety causal factors and quantify effects of the input variables to aviation safety. Then, taking the civil aviation safety data of an airline from 2017 to 2019: management factors, environmental factors, aircraft factors, human factors and external factors as the research object, the aviation safety causal prediction model is constructed based on random forest, and the importance analysis, model construction and prediction accuracy analysis of prediction variables are carried out. The results show that the random forest model could effectively predict the key factors of aviation safety and the changing trend of aviation safety, and the robustness and prediction performance are significantly improved from those of other models (support vector machine and artificial neural network model). In addition, the results of variable importance analysis show that environmental factors have the greatest impact on aviation safety from 2017 to 2019 and need to be controlled; on the contrary, management factors have the smallest impact on aviation safety and can be ignored.
Translated title of the contribution | Novel method of aviation safety causality prediction based on random forest |
---|---|
Original language | Chinese (Traditional) |
Pages (from-to) | 762-768 |
Number of pages | 7 |
Journal | Xi Tong Gong Cheng Yu Dian Zi Ji Shu/Systems Engineering and Electronics |
Volume | 45 |
Issue number | 3 |
DOIs | |
State | Published - Mar 2023 |