CAJ | 학술논문

监督学习情况下,经常遇到样例的维数远远大于样本个数的学习情况.此时,样例中存在许多与样例类标签无关的特征,研究如何同时实现稀疏特征选择并具有更好的分类性能的算法具有优势.提出了基于权核逻辑斯蒂非线性回归模型的分类和特征选择算法.权对角矩阵的对角元素在0到1之间取值,对角元素的取值作为学习参数由最优化过程确定,讨论了提出的快速轮转优化算法.提出的算法在十个实际数据集上进行了测试,实验结果显示,提出的分类算法与L1,L2,Lp 正则化逻辑斯蒂模型分类算法比较具有优势.
감독학습정황하,경상우도양례적유수원원대우양본개수적학습정황.차시,양례중존재허다여양례류표첨무관적특정,연구여하동시실현희소특정선택병구유경호적분류성능적산법구유우세.제출료기우권핵라집사체비선성회귀모형적분류화특정선택산법.권대각구진적대각원소재0도1지간취치,대각원소적취치작위학습삼수유최우화과정학정,토론료제출적쾌속륜전우화산법.제출적산법재십개실제수거집상진행료측시,실험결과현시,제출적분류산법여L1,L2,Lp 정칙화라집사체모형분류산법비교구유우세.
Under supervised learning settings, problems that the dimension of the samples is typically larger than the number of samples are often encountered, i.e. many irrelevant features exist. In such case, the approaches that simultaneously achieve sparsely variable selection and better accuracy of classification are more preferable. In this paper, classification and feature slec-tion algorithm based on kernel-weighted nonlinear logistic regression model is proposed. Each diagonal element of the weight diagonal matrix has a value between 0 and 1, which is as a learning parameter determined by optimization procedure, and fast alternative optimization methods are discussed. The proposed methods are tested on ten real-world datasets. The experimental results indicate that the proposed methods show high classification accuracies on these data sets than L1, L2, Lp norm regulariza-tion classifier algorithm of logistic regression model.