Under the background of digital education, the teaching management mode is shifting from teacher leading to student main participation, and the relationship between students ‘learning behavior and autonomic nervous system regulation needs to be objectively analyzed by computational methods. Aiming at the student-subject teaching management scenario, this paper integrates learning platform logs, classroom interaction records, heart rate variability, galvanic skin response and teaching feedback data, constructs multi-source state vectors, and designs feature normalization, autonomic nervous system regulation index calculation, state recognition and machine learning intervention decision-making methods. 120 undergraduates were included in the experiment, and 112 valid samples were finally retained, forming 16800 learning behavior records, 13440 physiological signal segments and 2860 teaching feedback records. The results showed that all dimensions of students ‘learning autonomy were significantly improved, RMSSD increased from 31.6 ms to 42.8 ms, and LF/HF decreased from 2.15 to 1.48. The AUC of the proposed model reaches 0.962, the F1 value reaches 94.1%, and the adjusted risk status is reduced from 16.0% to 5.4%. The research can provide technical support and practical basis for the monitoring of students ‘physical and mental state, intelligent teaching management and personalized intervention.
