Diagnosis and prediction of obesity and its effective factors using artificial neural network: A case study of children and adolescents residing in Isfahan-Iran

Overweight obesity is now so widespread in the world. This study aims to use an artificial neural network modeling tool to develop a predictive model for the diagnosis of obesity in children and adolescents.   Participants consisted of 460 school students, aged 7-18 years, who studied in a national school-based surveillance program (CASPIAN-V). Training network with 10 input variables including: age, sex, weight, height, waist circumference, systolic blood pressure, diastolic blood pressure, body mass index, waist-to-height ratio, physical activity, and with output variable obesity with 17 and 15 hidden neurons for girls and boys was designed.   After designing the network, the value of gradient on the data was 0.0021194 for girls and 0.0031658 for boys. The sensitivity, specificity and accuracy of the neural network were 0.9444, 0.9855, 0.9822, respectively in girls, and 0.9655, 0.9757, 0.9755 in boys;  in all these cases, the designed artificial neural network performed better than waist circumference and body mass index. A review of the final weights of this network showed that the input variable body mass index in girls and the input variable waist-to-height ratio in boys had the most influence in diagnosis of obesity.   Our results show that although body mass index has a better diagnostic performance in determining excess body fat than waist circumference, in boys and girls of both groups, and also in all parameters of sensitivity, specificity and accuracy, the artificial neural network acts better than body mass index and waist circumference, so that with an accuracy of more than 96%, we can detects obesity.