Search

Purpose: The use of aviation incident data to carry out aviation risk prediction is of great significance for improving the initiative of accident prevention and reducing the occurrence of accidents. Because of the nonlinearity and periodicity of incident data, it is challenging to achieve accurate predictions. Therefore, this paper aims to provide a new method for aviation risk prediction with high accuracy. Design/methodology/approach: This paper proposes a hybrid prediction model incorporating Prophet and long short-term memory (LSTM) network. The flight incident data are decomposed using Prophet to extract the feature components. Taking the decomposed time series as input, LSTM is employed for prediction and its output is used as the final prediction result. Findings: The data of Chinese civil aviation incidents from 2002 to 2021 are used for validation, and Prophet, LSTM and two other typical prediction models are selected for comparison. The experimental results demonstrate that the Prophet–LSTM model is more stable, with higher prediction accuracy and better applicability. Practical implications: This study can provide a new idea for aviation risk prediction and a scientific basis for aviation safety management. Originality/value: The innovation of this work comes from combining Prophet and LSTM to capture the periodic features and temporal dependencies of incidents, effectively improving prediction accuracy. [ABSTRACT FROM AUTHOR]