多维健康监测融合数据特征选择

Multidimensional Health Monitoring Fusion Data Feature Selection

中文关键词:  特征选择  特征空间  灰色关联度  融合

英文关键词:feature selection  feature space  gray correlation  fusion

中文摘要:

      健康监测通常使用大量传感器获取海量的感知数据，由于海量多维数据中存在大量的冗余或干扰，会对监测决策产生负面影响，为此需要对健康监测数据进行特征选择，旨在从数据中剔除多余的和不相关的特征。在现有研究的基础上，提出了特征选择融合方法，该方法通过ReliefF算法进行特征权重计算，并通过LASSO回归模型的计算结果确定特征权重阈值，进行特征初选，降低特征空间的稀疏性，然后利用灰色关联度的属性约简算法来消除冗余，从而获得最优特征子集。在实际多维感知数据集上进行测试，证明该模型可筛选出与目标参量相关性高的特征，降低回归运算的时间，提高回归模型的拟合精度。

英文摘要:

      Health monitoring usually uses a large number of sensors to obtain massive perceptual data.Because there is a large amount of redundancy or interference in the massive multidimensional data,which will have a negative impact on the monitoring decision,it is necessary to carry out feature selection on the health monitoring data to eliminate redundant and irrelevant features from the data.On the basis of the existing research,a feature selection fusion method is proposed.This method calculates the feature weight through the ReliefF algorithm,determines the feature weight threshold through the calculation results of the LASSO regression model,performs the feature selection,reduces the sparsity of the feature space,and then uses the attribute reduction algorithm of the grey correlation degree to eliminate the redundancy,so as to obtain the optimal feature subset.The test on the actual multi-dimensional perception data set shows that the model can filter out features with high correlation with the target parameters,reduce the time of regression operation,and improve the fitting accuracy of the regression model.