Takeover reliability during automated control transitions: An explainable analytical framework integrating scenario constraints and human performance shaping factors

In highly automated transport operations, operator recovery during automation-to-human control transitions is critical to mission continuity and system safety. Existing studies have not fully explained how takeover scenarios, human-related performance-shaping factors (HR-PSFs), and their coupling mechanisms affect takeover reliability. This study proposes an explainable human–system reliability framework integrating statistical inference, interpretable machine learning, global sensitivity analysis, and nonlinear threshold identification. Linear mixed models were used to examine the effects of takeover scenarios and HR-PSFs. A Tabular Prior-Data Fitted Network was then developed to predict takeover reliability, followed by Sobol sensitivity analysis, SHAP interpretation, and generalized additive modeling to identify key factors, interaction mechanisms, and operational thresholds. The results indicate that takeover trigger type was the dominant scenario factor, with unprompted pattern anomalies producing the most pronounced reliability degradation. External scenario conditions did not affect reliability uniformly. Instead, they amplified reliability fluctuations by changing task constraints and operator functional states. HR-PSFs did not act as independent linear predictors but formed context-dependent reliability-modification mechanisms and exhibited nonlinear threshold effects. These findings indicate that reliability degradation arises from dynamic mismatches among automation boundaries, operational contexts, and HR-PSFs. The proposed framework supports reliability monitoring, adaptive warning, and intervention design in safety-critical human–automation systems.