您的位置：首页 > 论文页面
基于混合策略的多标记学习算法及应用
发表时间：2017-05-31 浏览量：1812 下载量：672
| 全部作者： | 雷少正 |
| 作者单位： | 贵州民族大学数据科学与信息工程学院 |
| 摘 要： | 结合算法适应策略和问题转换策略，提出一种基于混合策略的多标记学习算法（multi-label learning based on hybrid strategy，HSML），其基本思想是k近邻（k-nearest neighbor，kNN）原则和k标记子集（k-labelsets，kEL）原则。该多标记学习框架首先重叠划分标记集，降低HSML预测标记的规模，然后提出基于分类准确率的子标记集排序算法，挑选一些子标记集，并针对每一个子标记集进行特征提取。在此基础上采用ML-kNN（multi-label learning based on k-nearest neighbor）算法对每一个子标记集进行分类，最后结合所有基分类器的结果对整个标记集进行分类。实验表明，对于局部标记相关的帕金森用药推荐数据集上，该算法性能优于ML-kNN和RAkEL（random k-label sets）算法。 |
| 关 键 词： | 人工智能；多标记学习；k近邻；k标记子集；局部标记相关性；帕金森用药推荐 |
| Title： | Multi-label learning algorithm based on hybrid strategy and its application |
| Author： | LEI Shaozheng |
| Organization： | School of Data Science and Information Engineering, Guizhou Minzu University |
| Abstract： | In this paper, a multi-label learning algorithm based on hybrid strategy (HSML) is proposed, combined with algorithm adaptation and problem switching strategy. Its basic thought is k-nearest neighbor (kNN) principle and k-labelsets (kEL) principle. The multi-label learning framework firstly overlaps the division of marker sets, and reduces the scale of predictive markers in HSML, then proposes a sub-label set sorting algorithm based on the classification accuracy, and selects some sub-label sets and performs feature extraction in each sub-label set. On this basis, an ML-kNN (multi-label learning based on k-nearest neighbor) algorithm is used to classify each sub-label set. Finally, the whole label set is classified according to the results of all the base classifiers. Experiments show that the proposed algorithm is superior to the ML-kNN and RAkEL (random k-label sets) algorithms for the data sets of Parkinson’s drug recommendation related to the local marker set. |
| Key words： | artificial intelligence; multi-label learning; k-nearest neighbor; k-label subsets; label correlations locally; Parkinson’s drug recommendation |
| 发表期数： | 2017年5月第10期 |
| 引用格式： | 雷少正. 基于混合策略的多标记学习算法及应用[J]. 中国科技论文在线精品论文，2017，10（10）：1117-1124. |
2
评论数 0请您登录
暂无评论
下载全文
多维论文