Application of Artificial Neural Network on Body Somatotype Analysis among Indian Population

Abstract

This study aims to verify the accuracy of the body somatotype analysis of the anthropometric measurements by using multi-layer perceptron Artificial Neural Network (MLP-ANN) model. It would also compare the predictive accuracy with a linear regression model using anthropometric body somatotype analysis (endomorphy, mesomorphy and ectomorphy) as reference method. A total of 293 persons (126 men and 167 women), between the ages of 19-40 years were recruited. The body somatotype was calculated from the anthropometric measurements using Heath-Carter technique. Linear regression equations and MLP-ANN prediction equations were developed. The variables - age, height, weight, humerus and femur breadth, arm and calf circumferences, subscapular, suprailiac, supraspinale and medial calf skinfold thicknesses, and sex as covariate, were used to develop the linear model (LR) to predict EndomorphyLR (coefficient of determination, R2=0.891; standard error of estimate (SEE) = 0.387), MesomorphyLR (coefficient of determination, R2=0.998; SEE = 0.068) and EctomorphyLR (coefficient of determination, R2=0.871; SEE = 0.553). The above variables were placed in the input layer of the MLP-ANN model (EndomorphyMLP, R2=0.909; MesomorphyMLP, R2=0.985; EctomorphyMLP, R2=0.994). The result reflects that MLP-ANN model had greater accuracy in predicting endomorphy and ectomorphy when compared to the linear model. However, linear model showed better accuracy for mesomorphy. Therefore, MLP-ANN model is more suitable in predicting body somatotype (endomorphy and ectomorphy) among population.