A Hybrid Learning-based Model for Predicting Diabetes using Tabular Network

Diabetes mellitus, a metabolic disorder characterized by hyperglycemia, poses a significant global health burden. Early and accurate diagnosis of diabetes subtypes is crucial for effective management and prevention of complications. This research addresses the challenge of multi-class diabetes type prediction using a local dataset. A deep learning approach, specifically Tabular Network, is employed to classify patients into different diabetes types. Tabular Network unique sequential attention mechanism dynamically selects relevant features at each decision step, enhancing model performance. Experimental results demonstrated the superiority of Tabular Network over other machine learning algorithms like Artificial Neural Network (ANN), Boosting and logistic regression. This research contributes to the advancement of diabetes diagnosis and offers a promising tool for healthcare professionals to make informed clinical decisions.