Research on Learning Behavior Prediction and Personalized Teaching in Smart Education Based on Big Data Analysis
Year: 2025
Type: conference-paper
Abstract: Vocational colleges often face two coupled problems: teachers cannot see in advance which students will fail key skill assessments, and the remedial content pushed to students is too generic to change outcomes. This paper presents a big-data-driven framework for learning behaviour prediction and personalized teaching in a real higher vocational setting. We collected multiyear logs from 21,000 students in eight popular majors, integrated LMS traces, practical-training records, assessment scores, and counselling records, and then carried out feature engineering to build a compact behaviour profile. On top of this profile, we designed a hybrid predictor that couples XGBoost for static and categorical features with a gated recurrent unit (GRU) network for time-series learning traces. The hybrid model achieved 93.2% early-warning accuracy for “skill assessment fail” cases and 88.6% accuracy for single skill points such as machining accuracy or e-commerce live-streaming operation. Based on the predictions, the system pushes different interventions (short video for weak practice items, skill-reinforcement modules mapped to target jobs, and task-level hints). In an A/B experiment, the intervention group improved the final skill-assessment pass rate by 18.5%, raised practical-task completion efficiency by 35.8%, and increased the average daily time spent on self-driven skill extension by 42.3 minutes compared with the control group. The results confirm that big-data learning analytics can front-load teaching interventions in vocational education and provide a practical pattern for smart campus deployment.
Institution: Chengdu Neusoft University
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: gold
OpenAlex ID: W7155387679