A comparison of multivariate statistical methods to detect risk factors for type 2 diabetes mellitus

Aim: The aim of this study was to compare the classification performance of Logistic Regression Analysis, Artificial Neural Networks and Decision Trees Methods by using data from patients with and without Type 2 Diabetes Mellitus and to determine risk factors for Type 2 Diabetes Mellitus. Material and Method: The data in this study were obtained from patients who came to İnönü University Faculty of Medicine Turgut Özal Medical Center Internal Medicine Department Diabetes and Thyroid Polyclinic. The data set consists of 25 independent variables and 1 dependent variable. Accuracy, sensitivity, specificity, precision, F-measurement, AUC and classification error were used in the performance criteria when comparing the classification performances of the methods. Results: Among the three methods, the best classification performance was given by the Artificial Neural Networks method. The accuracy, sensitivity, specificity, precision, F-measurement, AUC and classification error of this method were found as 98.94, 100, 97.73, 98.04, 99.01, 0.978 and 1.06, respectively. According to the results of Artificial Neural Networks method; from the risk factors affecting the disease, sex, family history, long-term drug use, cortisone use, concomitant disease, high blood pressure, stress factor, heart disease, high cholesterol, smoking, alcohol consumption, exercise status, carbohydrate use, vegetable use, meat use, age, weight , height, starting age, daily bread consumption, HDL, LDL, Triglyceride, Total Cholesterol, fasting blood sugar independent variables obtained weight values respectively; 0.017, 0.013, 0.009, 0.008, 0.017, 0.008, 0.016, 0.024, 0.053, 0.006, 0.007, 0.023, 0.040, 0.020, 0.007, 0.046, 0.083, 0.049, 0.024, 0.066, 0.083, 0.084, 0.031, 0.020, 0.244. Conclusion: When the Artificial Neural Networks, Logistic Regression and Decision Trees classification methods are applied, the Artificial Neural Networks method has shown the best performance and according to this method, the most important risk factor that may cause diabetes is fasting blood sugar.