Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (12): 2014-2027.DOI: 10.3778/j.issn.1673-9418.1912001

Previous Articles     Next Articles

DE-ELM-SSC+:Semi-supervised Classification Algorithm

PANG Jun, HUANG Heng, ZHANG Shou, SHU Zhiliang, ZHAO Yuhai   

  1. 1. School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan 430070, China
    2. Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System, Wuhan 430070, China
    3. School of Computer Science and Engineering, Northeastern University, Shenyang 110169, China
  • Online:2020-12-01 Published:2020-12-11

DE-ELM-SSC+半监督分类算法

庞俊黄恒张寿舒智梁赵宇海   

  1. 1. 武汉科技大学 计算机科学与技术学院,武汉 430070
    2. 智能信息处理与实时工业系统湖北省重点实验室,武汉 430070
    3. 东北大学 计算机科学与工程学院,沈阳 110169

Abstract:

The combinations of evolutionary algorithms (EA) and analytical methods have been extensively studied in the fields of machine learning in recent years. This paper focuses on how to combine a differential evolution (DE) algorithm with the semi-supervised classification algorithm based on extreme learning machine (ELM). Firstly, this paper proposes a semi-supervised classification algorithm based on DE and ELM (DE-ELM-SSC) with roughly three steps. Firstly, multiple differential evolution strategies are adopted to optimize the input weights and hidden biases of extreme learning machine, and an optimal strategy for the target data set is selected according to the root mean square error (RMSE). Secondly, the optimal evolutionary strategy selected in the previous step is applied to the DE algorithm to optimize the ELM network parameters. Thirdly, in order to construct a semi-supervised classification model, tri-training technology is used to realize the cooperative training of three improved ELM base classifiers. Then, a nonlinear method is adopted to improve the existing inertial strategy method and realize adaptive adjustment of scaling factor, so as to optimize the DE-ELM-SSC algorithm to obtain the DE-ELM-SSC+ algorithm. Lastly, a large number of experimental results on UCI data sets show that the DE-ELM-SSC+ algorithm outperforms the baseline methods with higher classification accuracy because of evolution strategy selection and improved scaling factor adaptive adjustment.

Key words: extreme learning machine (ELM), semi-supervised classification, strategy selection, differential evolution (DE), scaling factor

摘要:

演化算法和分析方法的结合是机器学习领域近几年的一个研究热点。研究如何将差分进化(DE)演化算法与基于超限学习机(ELM)的半监督分类算法相结合。首先,提出了一种基于DE和ELM的半监督分类方法(DE-ELM-SSC),该算法大致步骤为:采用多种差分进化策略对超限学习机输入权重和偏置参数进行优化,并根据均方根误差选出一个适合目标数据集的最优策略;将上一步选出的最优进化策略应用于DE算法,从而达到优化ELM网络参数的目的;为了构造半监督分类预测模型,采用Tri-training技术实现了三个改进ELM基分类器的协同训练。然后,采用非线性方法改进现有惯性策略方法,实现了缩放因子自适应调整,从而优化了DE-ELM-SSC算法,得到DE-ELM-SSC+算法。UCI标准数据集上的大量实验结果表明,DE-ELM-SSC+算法能根据数据集选择合适的进化策略,并自适应调整缩放因子,获得比Baseline方法更高的分类准确率。

关键词: 超限学习机, 半监督分类, 策略选择, 差分进化, 缩放因子