Developing a Deep Residual Network For The Early Identification of Adult-Onset Diabetes

Diabetes is a chronic metabolic disorder characterized by elevated blood glucose levels, resulting in progressive damage to essential organs such as the heart, blood vessels, eyes, kidneys, and nerves. Primarily affecting adults, type 2 diabetes, the most common form, emerges when the body exhibits resistance to insulin or fails to generate a sufficient quantity. The past three decades have witnessed a substantial rise in the prevalence of type 2 diabetes globally, spanning nations with diverse income levels. In the quest to improve patient care through the early detection of diabetes-related complications, this study employs machine learning (ML) models to identify and classify type 2 diabetes. The research evaluates the effectiveness of five ML models-K-nearest neighbor (K-NN), Bernoulli Naive Bayes (BNB), decision trees (DT), logistic regression (LR), and support vector machines (SVM)-as well as three deep learning models-Res-Net, MultiLayer Perceptron (MLP), and Long Short-Term Memory (LSTM)-in identifying individuals with diabetes.Using a Kaggle-hosted dataset that includes information from 2768 patients, both with and without diabetes, the study considers various factors such as previous pregnancies, blood glucose levels, diastolic blood pressure, skinfold thickness, body insulin levels, body mass index, genetic background, family history of diabetes, age, and ultimate outcome. The results indicate that the deep learning models demonstrated notable performance. The Res-Net model, with a success rate of $98 \%$, emerged as the most effective in diabetes identification, closely followed by the MLP and LSTM models with success rates of $\mathbf{9 4. 2 \%}$ and $87.2 \%$, respectively. In summary, the use of deep learning models for early diabetes detection shows significant promise, as evidenced by the favorable performance of the ResNet model in this investigation.