This study examined the effective use of modern artificial intelligence and deep learning-based neural network models, specifically MLP and TabNet models, for the early diagnosis of diabetes mellitus. The research focused on exploring efficient ways to utilize these neural network models in the early detection of the disease. In this study, TabNet and MLP models were selected and compared to automatically identify complex and nonlinear relationships in the medical dataset and to improve the effectiveness of diagnosis based on clinical indicators. The dataset underwent stages of preparation, cleaning, normalization, scaling, and feature selection through correlation analysis. The prepared dataset was split into 80% for training and 20% for testing. The methodological approach presented in this article opens up the possibility for widespread application in the automated diagnosis of other chronic and complex diseases in the future.