Accuracy Of Classifying Prediabetes Predicted By Grip Strength In Obese Adults Utilizing Machine Learning

• DOI: 10.1249/01.mss.0000678436.70420.41
• Publication Date: 2020-07-09T05:22:20Z
• Authors: Jaemyung Kim; Kwanghee Lee; Su-Hong Kim; Eunju Hong; Ji-Yeob Choi; Hyo Lee; Miyoung Lee
• Keywords: Prediabetes

PURPOSE: To compare the accuracy of classifying prediabetes, include type 2 diabetes, predicted by hand-grip strength employing machine learning (ML) techniques in Korean obese adults. METHODS: The data 1230 adults (51.5% males, 19-65yrs) was retrieved from National Health and Nutrition Examination Survey (KNHANES) 2014-2015. Obesity identified standard BMI (BMI ≥ 25kg/m2) waist circumference (WC; Male: WC 90cm, Female: 85cm). A total 591 individuals with prediabetes diabetes diagnosed criteria 1) as 2) using medication, 3) abnormal fasting glucose level (fasting 100mg/dL). Three grip models were employed, which normal (GS), divided weight (GSW), (GSB), respectively, to examine effect different relative strengths on prediabetes. Multilayer perceptron (MLP) traditional logistic regression (LR) algorithms RSNNS package R, applied classify adjusted age, income, education, occupation, marital status, binge drinking, smoking, daily calories intake, sedentary time, strengthening exercise adherence, aerobic adherence variables, separated gender. evaluate (ACC), sensitivity (SEN), specificity (SPE) confusion matrix ML, participants into a train for deriving equations test group holdout cross-validation. RESULTS: GSB co-variates revealed highest ACC both ML classifiers (train group: MLP=71.5%, LR=67.6%; MLP=63.9%, LR=66.1%), GSW showed lowest (MLP: train=67.6%, test=61.6%). Moreover, MLP higher SEN than SPE (SEN, SPE: GSW=78.6%, 44.1%; GSB=74.4%, 52.5%, respectively). In females, however, did not show any consistent levels. CONCLUSIONS: It that strong relationship male Korea. LR present fair cross-validation GSB.