Optimizing Timely Diabetes Detection Through Advanced Predictive Modeling and Patient Medical Statistics

Abstract

Diabetes has become prevalent across a significant portion of the global population. While medication offers one avenue for disease management, its dosage often escalates with age, eventually necessitating insulin injections. Alternatively, lifestyle adjustments provide another means of control. Detecting diabetes early holds the potential to enhance patient well-being and avert the onset of the disease. This study undertakes a thorough parametric analysis of diabetes, comparing Linear, Multi-Linear, Polynomial, and Logistic Regression Models. The goal is to identify the optimal approach for predicting diabetes. Leveraging a dataset encompassing 768 patients from the National Institute of Diabetes and Digestive and Kidney Diseases, USA, the research considers both Patient Medical Statistics (PMS) and a Combination of Patients Medical Statistics (CPMS), featuring variables such as Pregnancy, Blood Sugar (BS), Blood Pressure (BP), Body Fat (BF), Insulin Level (IL), Body Mass Index (BMI), Diabetes Pedigree Function (DPF), and Age. Evaluation metrics, including Root Mean Square Error (RMSE) and Coefficient of Determination (R2), guide model selection. By employing R2, the ranking of PMS and CPMS is determined to yield a highly efficient model. Empirical findings indicate the Multiple Linear Regression model as presenting the lowest RMSE among all regression models. However, the comparative assessment underscores the suitability of Logistic Regression due to its discrete nature, demonstrating superior accuracy in prediction compared to other models.