Risk prediction and early warning for air traffic controllers’ unsafe acts using association rule mining and random forest

Abstract

In the context of big data, scientific judgment of the future trend or state of unsafe acts of air traffic controllers (ATCers) plays an important role in the prevention of unsafe incidents. Therefore, based on the previous prediction and early warning studies, this study chose the methods of correlation analysis and nonlinear modeling to scientifically predict and warn ATCers’ unsafe acts. Firstly, according to the process of association rule mining, the association rules among risk factors and controllers' unsafe acts were mined. On the basis of association rule mining, the prediction and early warning model was established by random forest algorithm. The prediction data samples were used to predict ATCers' errors and violations respectively. The results show that the number of two-dimensional association rules is the largest, followed by one-dimensional and two-dimensional association rules, and four-dimensional association rules is the least. For the prediction of errors and violations, the deviation between the predicted value and actual value is small, indicating that the prediction accuracy is high. Business ability, technical environment, inadequate supervision and mental states are of greater importance to the prediction of errors, while inadequate supervision, organizational climate, organizational process and mental states are more important for the prediction of violations. The application of association rule and random forest model can effectively predict ATCers' errors and violations, and send signals to relevant departments according to the threshold and alarm level to report dangerous situations, thus reducing the losses caused by ATCers' unsafe acts.