Research On Human Abdominal Fat Area Prediction Model Based On Semi Supervised Learning

In order to predict the fat area(VFA)of human abdominal viscera(intraperitoneal)based on bioelectrical impedance,this dissertation uses the ABC-SVR prediction model based on semi supervised learning to predict the body fat area in human abdomen to overcome the problem that the training sample is limited and the correlation of the standard value is not high enough.The results show that the model has a strong nonlinear function approximation and can effectively predict the area of body fat in human abdomen.The main contents of this dissertation are as follows:1.The characteristic properties of predicting human intra-abdominal fat area is constructed.By using the human body's abdomen as an elliptical column conductor model,a measurement method of human abdominal bioelectrical impedance is given,and the characteristics associated with the crowd type are defined.The prediction model's feature of input of human abdominal fat area is constituted with eight characteristics:abdominal total impedance,abdominal subcutaneous impedance,waist circumference,abdominal width,abdominal thickness,body mass index B,body fat rate T,and waist-to-hip ratio Y.2.An ABC-SVR prediction model of human intra-abdominal fat area was constructed.Through the improved artificial bee colony algorithm,we perform global optimization training on the intra-abdominal fat area characteristics and find the optimal eigenvector solution.The optimal eigenvectors were fitted by supporting vector regression,and the ABC-SVR prediction model of intra-abdominal fat area was obtained.3.A semi-supervised learning algorithm for predicting intra-abdominal fat area was given.By optimizing labeling of newly collected unlabeled samples,the ABC-SVR prediction model is repeatedly trained on the added new marker samples and the original marker samples through the semi-supervised learning algorithm.Through comparison of performance by Akaike Information Standard with the established forecasting model,a new optimal forecast model is obtained to solve the problem that training samples can be continuously increased and the model can be repeatedly trained.4.The classical least squares regression model and ABC-SVR model are chosen as the comparison model for the semi-supervised learning ABC-SVR model.This method is feasible and effective through the prediction and comparative analysis of the test sample set.This dissertation uses Matlab to simulate the human abdominal fat area prediction model and test sample based on semi supervised learning.Through the analysis of correlation and standard deviation,the results show that the improved effect of this method is obvious.