A neural network-based model for cultivating applied talents in the context of industry-education integration

To address the demand for skilled applied professionals, this study proposes a neural network-based model for integrating industry and education. The model combines Transformer and Residual Network (ResNet) architectures and employs a data augmentation strategy. The study evaluates different teaching methods based on academic performance, practical skills, innovation, and professional competency alignment. The experiment used a dataset of student interactions from an online education platform. Comparative experiments assessed model performance under different data volumes. Results showed that the optimized model performed well with small datasets (1,000 data points), achieving a Log Loss of 0.106, a Mean Absolute Error (MAE) of 0.057, an Area Under the Precision-Recall Curve (AUC-PR) of 0.896, and an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.975. These results suggest that the model maintained predictive accuracy and stability with limited data. However, performance declined with larger datasets, where Log Loss increased to 0.160, MAE to 0.071, AUC-PR decreased to 0.87, and AUC-ROC to 0.961. This indicates the need for further optimization to improve robustness for large datasets. Compared to ResNet, the optimized model showed better performance with small datasets, while ResNet exhibited greater stability and lower Log Loss with larger datasets. The study also compared three teaching methods—Project-Based Learning (PBL), Flipped Classroom, and practical teaching. Results showed that practical teaching had the strongest impact on academic performance and practical skills. PBL and Flipped Classroom were more effective in fostering innovation and academic research skills. This study contributes to the optimization of teaching evaluation models and provides insights into improving education through neural networks. The findings may help refine talent cultivation models and teaching assessment methods.