Neuropathic complications: Type II diabetes mellitus and other risky parameters using machine learning algorithms

The World Health Organization (WHO) reported in 2016 that close to 422 million adults live with diabetes. Diabetes is described as the United Kingdom's fastest-growing disease, which in recent decades has almost doubled in prevalence (Zghebi et al., Diabetes Obes Metab 19(11):1537-1545, , 2019). The risk of DM is strongly advised, and the diabetes community should concentrate on identifying a pattern that leads to DM and its complications to avoid further complications. This study uses, logistic regression on PIMA Indian Diabetes Dataset, to identify the parameters and combination of parameters that lead to diabetic peripheral neuropathy (DPN) complications. Five different ML algorithms linear discriminant analysis (LDA), Naive Bayes (NB), C5.0, SVM_RBF (Support Vector Machine_Radial Basis Function and gradient boosting machine (GBM) has been applied for the diagnosis of such disease for enhance the identified models using LR. The prediction performances are evaluated with four different performance metrics that is viz., accuracy, kappa, AUC-ROC, MCC measures. The critical combinations include (1) DM + Glucose + BP + BMI + Age, (2) DM + Glucose + BMI + Age, (3) DM + Glucose + BP + BMI + Age, (4) DM + Glucose + BP + BMI, (5) DM + Glucose + BMI. For each method in this study, LR is used in DPN prediction as the Accuracy and AUC are more close to reality. The results after calculations are: 0.79, 0.77, 0.787, 0.772 and 0.769 respectively, the AUC are 0.83 for all models and the MCC are around 0.5. The prediction was tested with other machine learning methods such as LDA, NB, C5.0, SVM_RBF and GBM have been close to 0.77 and the Kappa is 0.4. In the investigation, the patients should be tracked, and doctors should be notified to prescribe medications to ensure them from further complications, if the behavior indicates either of these combinations.