A data-driven model of human factors contributing to loss of control in-flight and on the ground in general aviation
该研究利用美国NTSB历史数据，结合机器学习与动态贝叶斯网络，构建了一个量化人为因素对通用航空失控事件影响的模型，识别出技能型和决策型错误是主要诱因，并指出环境和心理前兆是最有效的干预目标。
Loss of control inflight and on the ground (LOC-I/G) is a leading contributor to fatalities in general aviation, often resulting from complex interactions between human, environmental, and systemic factors. This study presents a novel data-driven model that integrates machine learning, probabilistic forecasting, and Dynamic Bayesian Networks (DBNs) within a modified Human Factors Analysis and Classification System (HFACS) framework to quantify and forecast the influence of human errors on LOC-I/G incidents. Using historical data from the U.S. National Transportation Safety Board (NTSB), Random Forest (RF) and Support Vector Machine (SVM) classifiers were employed to estimate baseline probabilities of 12 major causal factors. These outputs were calibrated and combined with co-occurrence matrices to construct Conditional Probability Tables (CPTs), enabling the formation of a temporally responsive DBN. Forecasting capabilities were introduced via a Bayesian state-space model, and uncertainty was quantified using Monte Carlo simulations. The results reveal that skill-based and decision errors are the most prevalent contributors to LOC-I/G events, while backward sensitivity analysis identifies environmental and psychological precursors as the most effective targets for intervention. The proposed model provides dynamic risk evolution insights and supports real-time scenario analysis. Its modular design and reliance on empirical data make it adaptable to diverse aviation environments and regulatory settings. This research offers a practical, scalable framework for enhancing aviation safety through proactive, data-informed decision-making, with significant implications for training, supervision, and safety policy in both U.S. and global aviation systems.