Comparison of Logistic Regression and Support Vector Machine Models to Predict Diabetes

• Activity Number: 532
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Wei Yu*+ and Tiebin Liu and Rodolfo Valdez and Marta Gwinn and Nicole Dowling and Muin J. Khoury
• Companies: CDC and CDC and CDC and CDC and CDC and CDC
• Keywords: logistic regression (LR) ; support vector machine (SVM) ; National Health and Nutrition Examination Survey (NHANES) ; the receiver operating characteristic (ROC) ; body mass index (BMI)

Abstract

About 25% of the people with diabetes do not know they have the disease. We compared logistic regression (LR) and support vector machine (SVM) models in their ability to discriminate people with and without impaired glucose metabolism using the area under the ROC curve (AUC). We used data from the NHANES to develop and validate the models according to two classification schemes: I (diabetes vs. no diabetes) and II (diabetes or pre-diabetes vs. normal). Under scheme I, the best performing variables included family history, age, race or ethnicity, weight, height, waist circumference, BMI and hypertension. Under scheme II, sex and physical activity were added. Respectively, the AUC for the LR and SVM models were 83.5% and 83.2% for scheme I and 73.2% and 73.4% for scheme II. Both models appear to be promising classification tools for detecting diabetes and pre-diabetes in the US population.