APP下载

基于稀疏逻辑回归的链接模型在分类问题的应用

2021-07-20常鈺迪

软件工程 2021年6期

摘  要:随着科技的发展,数据分类问题应用在生活的多个方面,然而在面对庞大的数据时,往往采用压缩过的稀疏数据,这就为分类模型的发展带来了极大的挑战。为了提高稀疏数据分类的准确性和正确率,提出了基于稀疏逻辑回归的链接神经网络模型,由此构建成可靠的分类模型。以两类数据作为研究对象,首先进行数据预处理,再提取出数据特征对其进行分类。研究结果表明,分类模型不仅可以应用于稀疏数据,而且正确率较神经网络模型的结果有所提升,手写字的正确率从90.1%提高到94.86%,声音分类的正确率从70.3%提高到74.4%,证实该模型有效。

关键词:逻辑回归;稀疏性;神经网络;多分类

中图分类号:TP391     文献标识码:A

Abstract: With the development of science and technology, data classification is applied in many aspects of life. However, when facing huge data, compressed sparse data is often used, which brings great challenges to the development of classification models. In order to improve the precision and accuracy of sparse data classification, this paper proposes a link neural network model based on sparse logistic regression, so to build a reliable classification model. Taking two types of data as research object, data is preprocessed first, and then data features are extracted to classify them. The research results show that the classification model proposed in this paper can not only be applied to sparse data, but the accuracy is improved compared with the results of the neural network model. Accuracy of handwriting has increased from 90.1% to 94.86%, and accuracy of sound classification has increased from 70.3% to 74.4%, which proves that the model is effective.

Keywords: logistic regression; sparsity; neural network; multi-classification

1   引言(Introduction)

在現代数据分析中,具有挑战性的热点问题是从看似不足的数据量中恢复高维的信号,即数据的稀疏表示,这类问题在多个领域都有所涉及,例如压缩感知、稀疏近似和低秩矩阵恢复。本文受文献[1]1-Bit压缩感知中逻辑回归模型的收敛性以及可行性证明的启发,利用逻辑回归模型并结合稀疏性对实际应用问题进行研究。逻辑回归是最基本的回归形式,也是常用的分类方法。

现代生活的各个方面都离不开“分类”这一概念,应用逻辑回归模型解决分类问题备受研究者的关注,应用于图片分类[2]、医学诊断[3]等多个领域。本文对手写字和海洋哺乳动物数据集进行分类,海洋哺乳动物选取大西洋点斑原海豚、弓头鲸等10类作为研究对象,通过神经网络对分类数据集进行训练之后的训练集再进行逻辑回归模型训练的多分类实验,从而提高正确率。……

登录APP查看全文