Application of information-theoretic eye movement metrics in nuclear power Symptom-oriented emergency procedure training: Evaluation of training effects and cognitive load

Human errors contribute 60–90% to nuclear power plant major accidents. Symptom-Oriented Procedures (SOP) enhance diagnostic accuracy and response efficiency but elevate cognitive load due to multi-step logical judgments and extensive monitoring. Traditional metrics fail to capture real-time cognitive dynamics. This study pioneers the use of six information-theoretic eye movement entropy metrics to quantify gaze randomness, training effects, and load in SOP contexts. Eighteen novices performed repeated SOP tasks for steam generator tube rupture and small-break loss-of-coolant accidents on an M310 simulator. Unskilled and relatively skilled phases were compared through operation time, entropy, and NASA-TLX. Four entropy metrics (such as stationary gaze entropy, transition gaze entropy, dwell time entropy, and revisit entropy) decreased significantly with training (P < 0.05), and negatively correlating with improved performance and reduced load. A three-interval threshold (high: 0.72–1, medium: 0.04–0.72, low: 0–0.16) reflects strategy shifts from random exploration to targeted focus.