Evaluation methods based on data-driven techniques and artificial intelligence for the sustainable enrollment plan configurations of Chinese universities have become a research hotspot in the field of higher education teaching reform. Enrollment, education, and employment constitute the three key pillars of talent cultivation in universities. However, due to an unclear understanding of their interconnection, universities have yet to establish robust quantitative relationship models, hindering the formation of an evaluation mechanism for sustainable enrollment plan configurations. This study begins by constructing a relevant indicator system and utilizing real enrollment data from a specific university. Through statistical methods such as correlation analysis, it systematically sorts out key variables and identifies seven effective indicators, including average admission score and first-time graduation rate. Subsequently, by using the increase or decrease in enrollment quotas for each major as the experimental target, evaluation models for sustainable enrollment plan configurations aimed at enhancing the advanced education rate are constructed using naïve Bayes networks and tree-augmented Bayesian networks; these are compared with three other classic machine learning methods. The accuracy of these models is evaluated through confusion matrices and receiver operating characteristic curves. Additionally, the Birnbaum importance analysis method is utilized to prioritize remaining variables, ultimately identifying the optimal combination strategy of indicators conducive to the sustainable development of the advanced education rate. The results indicate that the average admission score, transfer rate, and student/teacher ratio are the top 3 prognostic factors affecting the advanced education rate, with the TAN model achieving an accuracy of 96.49%, thus demonstrating good reliability.

Accurately evaluating postgraduate education satisfaction is crucial for improving higher education quality and optimizing management practices. Traditional methods often fail to capture the complex behavioral interactions among influencing factors. In this study, an innovative satisfaction indicator system framework is proposed that integrates a two-stage feature optimization method and the Tree Augmented Naive Bayes (TAN) model. The framework is designed to assess key satisfaction drivers across seven dimensions: course quality, research projects, mentor guidance, mentor’s role, faculty management, academic enhancement, and quality development. Using data from 8903 valid responses, Confirmatory Factor Analysis (CFA) was conducted to validate the framework’s reliability. The two-stage feature optimization method, including statistical pre-screening and XGBoost-based recursive feature selection, refined 49 features to 29 core indicators. The TAN model was used to construct a causal network, revealing the dynamic relationships between factors shaping satisfaction. The model outperformed four common machine learning algorithms, achieving an AUC value of 91.01%. The Birnbaum importance metric was employed to quantify the contribution of each feature, revealing the critical roles of academic resilience, academic aspirations, dedication and service spirit, creative ability, academic standards, and independent academic research ability. This study offers management recommendations, including enhancing academic support, mentorship, and interdisciplinary learning. Its findings provide data-driven insights for optimizing key indicators and improving postgraduate education satisfaction, contributing to behavioral sciences by linking satisfaction to outcomes and practices.