Performance evaluation of artificial neural network and multiple linear regression in the prediction of body mass index in children

The body mass index (BMI) provides essential medical information related to body weight for the treatment and prognosis prediction of different diseases. The main goal of the present study was to evaluate the performance of artificial neural network (ANN) and multiple linear regression (MLR) model in the prediction of BMI in children. The data from a total of 5,964 children aged 5 to 12 years were included in study. Age, gender, neck circumference (NC), waist circumference (WC), hip circumference (HpC), and mid upper arm circumference (MUAC) measurements were used to estimate the BMI of children. The ANN and MLR were utilized to predict the BMI. The predictive performance of these methods was also evaluated. Gender-wise average comparison showed that median values of all the anthropometric measurements (except BMI) were significantly higher in boys as compared to girls. For the overall sample, the BMI prediction model was, [Formula] [-] 0.147 X Age [-] 0.367 X Gender + 0.176 X NC + 0.041 X WC + 0.060 X HpC + 0.404 X MUAC. A high R2 value and lower RMSE, MAPE, and MAD indicated that the ANN is the best method for predicting BMI in children. Our results confirm that the BMI of children can be predicted by using ANN and MLR regression methods. However, the ANN method has a higher predictive performance than MLR.