Reform of Learning Behavioral Characteristics Identification and Personalized Teaching Strategies in College Swimming Courses

To address the difficulty of identifying deep learning behaviors in high-capacity college swimming courses with large individual differences and large class sizes, this study proposes a multimodal data fusion and flexible leveled teaching reform scheme. Wearable sensors and video pose estimation are first used to collect students’ stroke frequency, stroke amplitude, breathing rhythm, heart rate variability, and movement posture data. A learning behavior recognition framework based on LSTM and K-Means++ is then constructed to classify students into three categories: technical proficiency, cognitive strategy, and basic imitation. Differentiated practice content, personalized load design, and real-time feedback are provided for each group through a mobile teaching assistant system. Experimental results show that the experimental group significantly outperforms the control group in swimming skill performance, with a score of 86.53 ± 4.67, and intrinsic motivation, with a score of 5.71 ± 0.43. The basic imitation group shows a 25.8-point advantage over the control group. The study provides a data-driven method for personalized swimming instruction and dynamic teaching adjustment.