Predicting Diabetes Using Stacked Dense Forwarded Network Model with Variable Analysis

The prevalence of diabetes, a chronic and potentially life-threatening illness, continues to rise globally. Each year, it contributes to a significant number of fatalities. Alarmingly, many individuals remain unaware of the severity of their condition until it has progressed to an advanced stage. To address this, the development of effective early detection methods is imperative, as delayed diagnosis often leads to serious health complications and increased mortality. This study introduces a deep learning (DL) framework aimed at predicting the likelihood of diabetes in individuals. The proposed method leverages a Stacked Dense Forwarded Network Model (SDFNM) for classification, integrating variable augmentation to enhance predictive performance. The model utilizes key patient attributes from the Pima Indians Diabetes Dataset (PIDD), including blood pressure, age, number of pregnancies, and insulin or glucose levels. The simultaneous training of the SDFNM classifier and the variable augmentation mechanism on a balanced dataset yielded a remarkable accuracy of 98%, marking a 4% improvement over existing approaches.