A Machine Learning Approach to Diabetes Diagnosis: Evaluating Logistic Regression and SVM Models on NHANES Data

Abstract

The goal of this study is to examine the trends of type 2 diabetes prevalence and the impacts of various demographic and socioeconomic variables on diabetes in the US. A top-down approach was used to develop logistic regression models for Type II diabetes diagnosis and for pre/undiagnosed Type II diabetes diagnosis. From these models, the most significant patient demographics were determined. Support vector machines (SVMs) were then trained and tested to compare the efficacy of the logistic regression models, and it was determined that the top-down logistic regression models consistently performed just as well as or better than SVMs. Gender, age, BMI, waist circumference, and race were all found to have strong impacts on the risks of developing Type II diabetes.