Risk early warning for unmanned aerial vehicle operators' unsafe acts: A prediction model using Human Factors Analysis and Classification System and random forest.

Abstract

The prediction of unmanned aerial vehicle (UAV) operators' unsafe acts is critical for preventing UAV incidents. However, there is a lack of research specifically focusing on UAV operators' unsafe acts, and existing approaches in related areas often lack precision and effectiveness. To address this, we propose a hybrid approach that combines the Human Factors Analysis and Classification System (HFACS) with random forest (RF) to predict and warn against UAV operators' unsafe acts. Initially, we introduce an improved HFACS framework to identify risk factors influencing the unsafe acts. Subsequently, we utilize the adaptive synthetic sampling algorithm (ADASYN) to rectify the imbalance in the dataset. The RF model is then used to construct a risk prediction and early warning model, as well as to identify critical risk factors associated with the unsafe acts. The results obtained through the improved HFACS framework reveal 33 risk factors, encompassing environmental influences, industry influences, unsafe supervision, and operators' states, contributing to the unsafe acts. The RF model demonstrates a significant improvement in prediction performance after applying ADASYN. The critical risk factors associated with the unsafe acts are identified as weak safety awareness, allowing unauthorized flight activities, lack of legal awareness, lack of supervision system, and obstacles. The findings of this study can assist policymakers in formulating effective measures to mitigate incidents resulting from UAV operators' unsafe acts.