Abstract: General aviation has developed rapidly in recent years, but the resulting safety problems attracted increasing attention. However, due to the wide variety of general aircraft and great difference between the samples, the traditional statistical analysis technology in general aviation risk forecast seems powerless. In this paper, a new prediction method based on neural network of sparse de-noising auto-encoder (SDAE) is proposed. SDAE can learn relatively sparse and concise data features and express input data better. By collecting the total number of other unsafe events from the time of January 2012 to December 2015 for 48 months, the neural network forecasting model of the civil incident rate is established to associate the occurrence of other unsafe incidents with the incident symptom. Examples show that the SDAE model can accurately predict the number of incidents in the month based on the number of other unsafe events.