Research on Strategic Adjustment Mechanisms of University Talent Cultivation Driven by Intelligent Learning Data

Learning traces generated through digitalization of higher education on a massive scale may promote responsive talent cultivation. This paper presents a closed-loop strategic adjustment mechanism that consists of four elements including learner profiling, explainable risk prediction, strategy matching, and result evaluation. Data for the 480 learners enrolled in six required courses were collected through Moodle activity logs, academic records, and a 20-item competency test. K-means analysis revealed four learner profiles, and XGBoost was used for predicting learning risk and competency gap with the help of behavior and academic performance in the first half semester. Subsequently, SHAP explanations were transformed into four intervention strategies that were practiced for eight weeks. In terms of the manuscript simulation results, learning risk prediction model demonstrated AUC score 0.906 ± 0.017 and F1 score 0.842 ± 0.031. Specifically, high-risk and low-engaged students increased their academic performance by 58.9 ± 7.6 to 68.7 ± 7.2, whereas the probability of being high risk decreased to 0.389 ± 0.162 from 0.713 ± 0.151. It can be concluded that integration of profiling, explainable prediction, and strategy matching can transform learning data into the tool of talent cultivation.