【摘要】：針對肝病分類中存在的特征交互的問題,我們研究了一種分層交互lasso分類方法。首先對logistic模型添加lasso罰函數(shù)和分層凸約束,其次采用卡羅需-庫恩-塔克條件與廣義梯度下降法相結(jié)合的凸優(yōu)化方法給出模型求解方法,最后得到主效應(yīng)特征系數(shù)與交互特征系數(shù)的稀疏解,實現(xiàn)模型分類。本文在兩個肝病數(shù)據(jù)集上進(jìn)行實驗,證明了特征交互對肝病分類有貢獻(xiàn)。實驗結(jié)果證明了分層交互lasso方法可解釋性強,效果、效率均優(yōu)于lasso方法、全特征對lasso方法以及支持向量機、最近鄰和決策樹等傳統(tǒng)分類方法。
[Abstract]:In order to solve the problem of the character interaction in the classification of liver disease, we have studied a hierarchical interactive lasso classification method. In this paper, the lasso penalty function and the hierarchical convex constraint are added to the logistic model, and then the method of the model is given by the convex optimization method which is combined with the generalized gradient descent method by the Caro-Kuhn-Tucker condition and the generalized gradient descent method, and finally, the sparse solution of the main effect characteristic coefficient and the interaction characteristic coefficient is obtained, and the model classification is realized. In this paper, we experiment on the data set of two liver diseases, and it is proved that the characteristic interaction contributes to the classification of liver diseases. The experimental results show that the layered interactive lasso method can be interpreted as strong, the effect and the efficiency are better than that of the lasso method, the whole characteristic is the lasso method, the support vector machine, the nearest neighbor and the decision tree, and the like.
【作者單位】： 燕山大學(xué)信息科學(xué)與工程學(xué)院;
【基金】：國家自然科學(xué)基金資助項目(61473339) 中國博士后科學(xué)基金資助項目(2014M561202) 河北省2014年度博士后專項資助項目(B2014010005) 首批“河北省青年拔尖人才”資助項目
【分類號】：R575
，

本文編號：2445689