Improved Grey Markov Aviation Safety Prediction Based on XGBoost

The number of accident indicators and accidents are critical metrics reflecting aviation safety. The Grey-Markov prediction model effectively forecasts aviation safety metrics such as accident indicators and accidents, especially in scenarios with limited samples, nonlinearity, and randomness. It has been demonstrated that different Grey-Markov models exhibit varying prediction performances depending on the data. Therefore, this paper explores an improved Grey-Markov model based on XGBoost for intelligent state partitioning, aiming to enhance the robustness and accuracy of the Grey-Markov model. Four sets of aviation safety data were used for empirical validation, yielding an average relative error of 0.0973. Compared to traditional Grey-Markov models with a single state partition, this approach proved to be more intelligent and accurate, showcasing its potential in predicting small sample, nonlinear time series data. This model offers valuable insights for aviation managers in macro-level safety management.