Table 4. Studies investigating prediction of diabetes using machine learning methods.
| Reference | Method | Predictors | Sample Size | Type of prediction | Performance |
|---|---|---|---|---|---|
| Yu et al. 2010 | SVM | family history, age, gender, race and ethnicity, weight, height, waist circumference, BMI, hypertension, physical activity, smoking, alcohol use, education, and household income(NHANES Cohort). | 4915 | Cross-sectional | AUC = 0.73 |
| Mani et al. 2012 | RF | A1c,Sys BP,Diastolic BP, GLU, BMI, Creatinine, HDL, MDRD, Triglycerides, Race, Gender, Age(EHR Data). | 2280 | 1 year ahead | AUC = 0.80 |
| Choi et al. 2014 | SVMANN | age, body mass index, hypertension, gender, daily alcohol intake, and waist circumference(KNHANES cohort) | 4685 | Cross-sectional | AUC = 0.74 |
| Anderson et al. 2016 | age,gender,systolic/diastolic BP, Height, Wieght, BMI, 150 ICD9 code, 150 common meds(HER data). | 9948 | Cross-sectional | AUC = 0.81 | |
| Luo 2016 | BRT + RF | The data set includes information ondemographics, diagnoses, allergies, immunizations, lab results, medications, smoking status, and vital signs. | 9948 | 1 year ahead | Accuracy = 87.4% |
| Our Study | RF15 | Hemoglobin A1c, fasting glucose, waist circumference, adiponectin, BMI, hs-CRP, triglycerides, age, leptin, body surface area, eGFR, 2D calculated left ventricular mass, HFL cholesterol, LDL cholesterol, aldosterone. | 3633 | 8 years ahead | AUC = 0.82Accuracy = 75% |
ANN–Artificial Neural Networks; BRT +RF–Combination of Boosting Regression Trees and RF classifiers.