Generative adversarial networks for student growth prediction in industry-education collaborative education

This study presents a teaching quality evaluation system (TQES) with a B/S architecture, integrating role-based access for administrators, teachers, and students to advance smart education through adaptive, data-driven decision-making. A GAN-based model predicts student growth using temporal academic data, including scores, feedback, and teaching metrics, addressing the need for dynamic and personalized trajectory modeling in modern pedagogical ecosystems. The Generator and Discriminator are optimized via adversarial training, minimizing MAE and RMSE through iterative gradient descent, while preprocessing ensures temporal coherence through normalization, missing-data imputation, and segmentation. Experiments on longitudinal data from 15,678 students (2018–2023) demonstrate the GAN’s superiority over traditional models such as Linear Regression, SVM, and Random Forest, achieving optimal performance (MAE: 0.25; RMSE: 0.5) at 2,500 iterations. Larger datasets enhance accuracy, with diminishing returns beyond 10,000 points. By aligning with smart education principles such as multimodal data fusion and real-time feedback, the model enables proactive intervention in industry-education collaboration. The architecture extends scalable solutions for adaptive pedagogical optimization, transforming static evaluations into predictive intelligence and fostering explainable, learner-centric analytics in smart education ecosystems.