Machine Learning Methods to Predict Incidence Risk of Altitude Decompression Sickness

Altitude decompression sickness (ADS) is a special disease when pilots are exposed to a low-pressure environment during training or on duty, which could endanger lives in severe cases. Accurate prediction of the incidence rate of decompression sickness is a major direction for diagnosis and treatment. At present, there are models such as bubble dynamics models and survival models based on logarithmic logistic distribution, but they heavily rely on the validity of the relationship among variables, resulting in limited predictive ability. In this paper, two machine learning methods, namely random forest regression and neural network, were used to train data from the high altitude decompression sickness risk assessment computer (ADRAC) of the US Air Force Research Laboratory and then predict the incidence of ADS under common missions. The results showed the mean square error of the random forest regression model on test data was 0.08958, the mean average absolute error was 0.01885 and the explanatory variance reached 0.99. When verified against real data proposed in previous literature, random forest regression performed better than ADRAC and neural network model, providing a feasible approach to predict the incidence rate of altitude decompression sickness using machine learning methods.