Research on Diabetes Analysis Based on Deep Learning-Enhanced Data

Diabetes poses a global health challenge, with accurate classification and blood glucose prediction being crucial for effective management and treatment. However, traditional methods are hindered by data scarcity. This study aims to use deep learning to complete data expansion to improve the accuracy of diabetes classification and blood glucose prediction. Raman spectroscopy, Conditional Generative Adversarial Networks (CGAN), and a suite of optimized regression models were employed, with the best hyperparameter combinations for each model on the current dataset determined through grid search and cross-validation. The introduction of CGAN for data augmentation effectively addressed the issue of data scarcity, resulting in a significant improvement in overall performance. The top-performing model achieved a diabetes classification accuracy of over 99% and a blood glucose prediction error of 0.28 mg/dL. The application of CGAN notably enhanced both classification and prediction accuracy, a trend further supported by performance improvements illustrated through comparative graphs. This integrated approach provides a robust solution for accurate diabetes diagnosis and blood glucose prediction, demonstrating potential advancements in non-invasive diabetes management.