摘要
基于变量交互和分层思想,提出了一种弱分层交互Lasso罚logistic回归模型。首先给出了交互模型定义和弱分层约束条件,然后给出了凸松弛条件和基于坐标下降法的系数求解算法。在4个UCI机器学习数据集和1个日常生活活动识别数据集上进行实验,实验结果证明了变量交互对分类也有贡献,分层对分类也有贡献。分层交互Lasso兼具Lasso和交互Lasso的优点。
Based on hierarchy theory and variable interactions,the hierarchical interactive Lasso penalized logistic regression model is proposed.First,an interaction model and the hierarchical constraint conditions are defined,and then the coordinate descent algorithms are used to solve model coefficients.The experiments on four UCI data sets and activities of daily living recognition are performed.The experimental results prove that the variable interaction makes great contribution to the classification results,and hierarchy contributes to the classification results.Hierarchical interactive Lasso shows more advantages than the Lasso and interactive Lasso.
作者
李静
于辉
王金甲
Li Jing;Yu Hui;Wang Jinjia(School of Science, Yanshan University, Qinhuangdao 066004;School of Information Science and Engineering, Hebei Provincial Key Laboratory of Information Transmission and Signal Processing, Yanshan University, Qinhuangdao 066004)
出处
《高技术通讯》
EI
CAS
北大核心
2020年第4期348-357,共10页
Chinese High Technology Letters
基金
国家自然科学基金(61473339,61771420,61501397,81803958)
京津冀基础研究合作专项(19JCZDJC65600,F2019203583)
燕山大学青年教师自主研究计划课题(15LGA015)
燕山大学博士基金资助项目。
