Enhancing Student Retention in Chinese Universities through Intelligent Predictive Modeling of Dropout and Graduation Delays

HRMARS - The research at hand explores the use of intelligent predictive modeling for solving the increasing alarm of dropout and time-to-degree in undergraduate student populations in Chinese higher education. By utilizing improved decision tree algorithms, including a new Random Forest–Neural Network (RF-NN) hybrid model, the research investigates a diverse dataset consisting of demographic, academic, financial, psychological, and institutional factors. The quantitative part uses supervised machine learning methods, but enhanced by Feature Engineering and by data balancing methods like SMORE, so that the model reliability is maximized. Model performance was evaluated using metrics including but not limited to accuracy, precision, recall, F1-score, and AUC-ROC. Academic failure, financial hardships, mental health problems, and practical difficulties are the strongest predictors of dropout and enrollment prolongation. In particular, the inclusion of culturally specific constructs, namely guanxi, Confucian style of pedagogical structures and the dual-teacher model, strongly aligned with Chinese education settings and further improved the contextual fit of the model. This research is novel in its approach to advance educational data mining and provide stakeholders with useful insights which can help institutional leaders to formulate proactive and data-informed early-intervention strategies. By identifying at-risk students and implementing appropriate and timely policies, the proposed predictive framework can be used to reduce dropout rates and facilitate on-time graduation. Researchers could build on this work by evaluating integrations with real-time analytics or scalability of this approach across different institutional contexts in China.