基于交替方向乘子法的广义交互LASSO模型用于肝脏疾病分类

发布时间：2018-02-24 09:09

  本文关键词： 肝脏疾病分类 特征交互 最小绝对收缩和选择算子 逻辑斯特回归 交替方向乘子法　出处：《生物医学工程学杂志》2017年03期 　论文类型：期刊论文

【摘要】：肝脏疾病特征及交互特征对于肝脏疾病的分类具有重要意义,本文在交互最小绝对收缩和选择算子(LASSO)模型的基础上,研究了广义交互LASSO模型并与其他可用于肝脏疾病分类的方法比较。首先,本文建立了广义交互逻辑斯特(logistic)分类模型,在模型参数中添加LASSO罚函数,然后将模型参数通过交替方向乘子法(ADMM)求解,得到模型系数的稀疏解。最后将测试样本代入模型,按照最大概率进行分类结果统计。通过将本文方法应用在肝脏失调数据集和印度肝病数据集的数据实验结果表明,交互特征的模型系数不为零,这说明交互特征对分类存在贡献。最终结果表明,本文提出的广义交互LASSO方法的正确率要优于交互LASSO方法,也优于传统模式识别方法,可将广义交互LASSO方法推广应用到其他疾病的分类问题上。
[Abstract]:The characteristics of liver diseases and interactive features are of great significance to the classification of liver diseases. This paper is based on the interactive minimal absolute contraction and selection operator LASSO model. The generalized interactive LASSO model is studied and compared with other methods that can be used in the classification of liver diseases. Firstly, the generalized interactive logic logistic classification model is established, and the LASSO penalty function is added to the parameters of the model. Then the model parameters are solved by alternating direction multiplier method (ADMMM), and the sparse solution of the model coefficients is obtained. Finally, the test samples are substituted into the model. According to the maximum probability of classification, the experimental results show that the model coefficient of the interactive feature is not zero, by applying this method to the liver maladjusted data set and the Indian liver disease data set. The result shows that the generalized interactive LASSO method proposed in this paper is better than the interactive LASSO method and the traditional pattern recognition method. The generalized interactive LASSO method can be extended to the classification of other diseases.
【作者单位】： 燕山大学理学院;燕山大学信息科学与工程学院;
【基金】：国家自然科学基金(61273019,61473339) 河北省青年拔尖人才支持计划([2013]17) 中国博士后科学基金面上项目(2014M561202) 燕山大学青年教师自主研究计划课题(15LGA015)
【分类号】：O212.1;R575
，

本文编号：1529662