Early Detection of Diabetes Using Machine Learning Algorithms: A Comparative Study

Until serious complications arise, diabetes is often overlooked as one of the most pressing health issues today. By examining a vast collection of clinical records, we explore how machine learning can be utilized to predict early diabetes outcomes. A database of 100,000 patient records is available, containing information on their age range, BMI and other medical conditions, as well as their smoking history and current medical problems. Our experiment used several classical models based on Machine Learning for instance Logistic Regression, Random Forest, Decision Tree, Naive Bayes, K-Nearest Neighbors, and Support Vector Machine, and it has been tested out on our deep learning model FUGA-NET, which stands for the combination of Feature Utility with Grouping and Adaptive fusion Network. In contrast, FUGA-NET utilizes adaptive modeling techniques that combine features to improve prediction. The conventional models were able to outperform Random Forest with 97.01% accuracy, while FUGA-NET achieved a higher ROC AUC of 0.9753. The results indicate that healthcare tools can use deep learning to improve the speed and reliability of diabetes screening.