Research on college physics learning behavior data modeling and teaching optimization based on SOLO classification theory

Abstract

The fields of learning analytics and educational big data are experiencing rapid development. Research in university physics education focuses on leveraging student behavioral data for precision instruction. This study employed the SOLO (Structure of Observed Learning Outcomes) classification model, developing a data model for university physics learning behavior involving feature extraction, hierarchical label mapping, classification model construction, and trajectory tracking. The overall system framework encompasses data collection, feature management, model training and real-time inference, an instructional intervention system, and a feedback loop. This system can predict individual SOLO levels and analyze group-level migration dynamics. Based on the 2024 Streaming Learning Analytics public dataset, the data was cleaned and aggregated weekly, yielding 1.7 million samples. During the classification prediction phase, the proposed random forest model achieved a weighted F1 score of 0.81 and a Cohen's κ score of 0.69, with predictions highly consistent with teacher evaluations. Subsequent teaching intervention experiments further revealed that after adopting the stratification strategy, the final grades of the experimental group students increased by an average of 4.2 points, and their learning satisfaction and independent learning ability were significantly improved. Analysis of the trajectory found that low-level students gradually transitioned, while high-level students achieved substantial improvement. Combining SOLO classification theory with computer modeling can significantly optimize the personalized process of university physics teaching and provide empirical data for the digital reform of education.