The Application of Artificial Neural Network on the Prediction of Human Body Dimensions

In this paper, the objective was the development of a genetic algorithm optimized generalized regression neural network (GRNN) for hum an body dimensions prediction. The GRNN prediction results were compared to those obtained with the standard multiple linear regression (MLR). 16 human body dimensions of every measured student obtained from non-contact measurement system (Symcad) were divided into control parameters and detail parameters for clothes, which corresponding to input and output units of the GRNN and independent and dependent variables of the MLR model. The training and test samples were used to establish and test the performance of the GRNN and MLR model. The error function for evaluating the quality of GRNN and MLR was the mean square error of the training set by leave-one-out cross-validation. The genetic algorithm taking the mean square error as the fitness function was used to optimize the smoothing factor of the GRNN. To evaluate the prediction accuracy of the two models, coefficient of correlation, mean square error and average absolute error are taken as the performance criteria. The optimized GRNN outperformed the MLR in terms of the prediction accuracy, even though it required more computational time than MLR.