In the context of the flourishing development of artificial intelligence in the education field, the improvement of normal students’ educational practice ability through AI - based means has become an urgent issue. This study developed an AI - supported”three-step” training model for normal students. It uses the K - means clustering algorithm (k = 5) for learning style classification and personalized learning path creation. BiLSTM - attention networks (3 - layer, dropout = 0.3) predict teaching behavior (F1 - score = 0.87), and a BERT - based NLP model (fine - tuned with 12,000 examples) provides 91.3% accurate real - time feedback. A virtual lab (89.2% task completion rate) is integrated into the training. In the six-month controlled experiment (N=350), compared with the traditional method, the”three-step” model achieved a significant improvement in the teaching ability score, with an overall improvement of 28.4% (P < 0.001, η = 0.42), especially in the aspects of classroom management and teaching design, which increased by 37.1% and 29.6% respectively. In addition, in another group of experiments (n=200), the evaluation score of TPACK is verified by A/B test, which shows that the model can improve the teaching efficiency by 23.7%(p<0.05, Cohen’s d=1.2). These results show that the AI-driven training model, which combines machine learning and natural language processing technology, has a significant effect on improving the comprehensive ability of teacher candidates. © 2025 Copyright held by the owner/author(s).
Author Keywords
Index Keywords
