基于半监督学习的人体腹部脂肪面积预测模型

PREDICTION MODEL OF HUMAN ABDOMINAL FAT AREA BASED ON SEMI-SUPERVISED LEARNING

人体腹部脂肪面积与人体腹部生物电阻抗和腹部皮下阻抗测量值存在着强非线性关系。采用人工蜂群算法对标记样本输入特征属性寻优获取最优解,通过支持向量回归机进行建模,计算赤池信息量准则(AIC)取得最优多项式模型。半监督学习方法通过反复预测大量未标记样本集,组建新的输入标记样本集进行重新寻优、训练、预测,计算最优AIC值的预测模型。通过测试样本集的标准差和相关性仿真实验表明,该模型具有较强的非线性函数逼近,能有效预测人体腹部脂肪面积。

There is a strong nonlinear relationship between body abdominal fat area and human abdominal bioelectrical impedance and abdominal subcutaneous impedance measurements. The artificial bee colony algorithm is used to get the optimal solution for the marking sample input feature attribute optimization. Through support vector machine regression modeling, we calculated the AIC（Akaike information criterion） to get the best polynomial model. Semi-supervised learning method repeatedly predicted a large number of unlabeled samples, and set up new input marker sets to reoptimize, train and predict, and calculated the best AIC value prediction model. By testing the standard deviation and correlation of the sample set, the simulation experiment shows that the model has strong nonlinear function approximation, and it can predict the abdominal fat area effectively.