Prediabetes Prediction Using Response Surface Methodology and Probabilistic Neural Networks Model in an Ethnic South Indian Population

Prevalence of prediabetes is increasing in India. Epidemiological studies have reported a strong association of anthropometric and biochemical parameters with the glycemic levels of prediabetes subjects. Identification of predictors of prediabetes using statistical and neural network models has been reported in this chapter. This chapter included 300 subjects from Timiri Block, Vellore, Tamilnadu, India, for neural network modeling. The subjects’ biochemical and anthropometric variables are analyzed, and the glycemic levels of samples are predicted. Pearson correlation analysis and response surface method are performed using Minitab software. Artificial neural networks and probabilistic neural networks are served using Matlab software. Results obtained in all the models are validated using a testing data set. It shows an accuracy of 95% for the probabilistic neural network model. Salivary glucose, HbA1C, waist circumference, BMI, and LDL are found to be predictors of prediabetes.