摘要: |
为提高辐射源个体识别的准确率,降低动态选择性集成的计算复杂度,提出基于特征选择和聚类的动态选择性集成学习模型。利用归一化皮尔森相关系数法度量基分类器间混淆矩阵的差异性,以各基分类器准确率最高及基分类器间差异性最大为目标,得到基分类器集合和对应特征子集集合。利用聚类方法将验证集划分为若干类,以验证集分类准确率最高为目标,为每簇验证集选择最优的基分类器子集和对应的特征子集。在测试阶段,对测试集进行聚类,仅比较每簇测试样本和每簇验证样本数据分布的最大均值差异值,减少运算时间。每簇测试样本在相似度最高的验证集所对应的特征子集集合和基分类器子集下进行预测,并根据不同权重基分类器预测结果的加权和进行最终决策。为验证方法的必要性和优越性,将本文方法与传统集成学习方法进行对比,结果表明,本文方法在信噪比分别为10dB、5dB条件下,分类准确率均提升约5%,具有更好的分类效果和泛化性能。 |
关键词: 特征选择 动态选择性集成 支持向量机 蚁群算法 辐射源个体识别 二分类问题 |
DOI: |
投稿时间:2022-10-06修订日期:2024-10-18 |
基金项目:国家自然科学(61371196);中国博士后科学基金特别资助项目(2015M582832);国家重大科技专项(2015ZX01040201-003) |
|
Dynamic Ensemble Selection based on Feature Selection and Clustering |
Xu Yuxin1, Cao Jianjun1, Wang Baowei2, Weng Nianfeng1, Gu Chumei1
|
(1.The rd Research Institute,National University of Defense Technology,Nanjing,Jiangsu;2.College of Computer Science and Technology,Nanjing University of Information Science and Technology,Nanjing,Jiangsu) |
Abstract: |
In order to improve the accuracy of individual recognition of radiation sources and reduce the computational complexity of dynamic selective integration, a dynamic selective integration learning model based on feature selection and clustering is proposed. The normalized pearson correlation coefficient method is used to measure the difference of the confusion matrix between the basis classifiers, and the base classifier set and the corresponding feature subset set are obtained with the goal of maximizing the accuracy of each base classifier and the difference between the base classifiers. The verification set is divided into several classes by clustering method. Aiming at the highest classification accuracy of the verification set, the optimal subset of base classifiers and corresponding feature subsets are selected for each cluster of verification set. In the progress of testing, the test set is clustered, and only the maximum mean difference value of the data distribution of each cluster of test samples and each cluster of validation samples is compared to reduce the operation time. Each cluster of test samples is predicted under the feature subset set and the base classifier subset corresponding to the verification set with the highest similarity, and the final decision is made according to the weighted sum of the prediction results of different weight base classifiers. In order to verify the necessity and superiority of the method, this method is compared with the traditional ensemble learning method. The results show that the classification accuracy of this method is improved by about 5% on average when the signal-to-noise ratio is 10dB and 5dB respectively, and it has better classification effect and generalization performance. |
Key words: feature selection dynamic ensemble selection support vector machine ant colony algorithm specific emitter identification binary classification |