选择性集成极限学习机分类器建模研究

RESEARCH ON MODELLING SELECTIVE ENSEMBLE EXTREME LEARNING MACHINE CLASSIFIER

极限学习机ELM(Extreme Learning Machine)具有训练过程极为快速的优点,但在实际分类应用中ELM分类器的分类精度和稳定性有时并不能满足要求。针对这一问题,在ELM用于分类时引入一种训练结果信息量评价指标来改进输出权值矩阵的求解方法,并增加隐层输出矩阵竞争机制来提高ELM的稳定性。为了进一步提高ELM的分类正确率,借鉴神经网络集成的理论,提出一种选择性集成ELM分类器。在集成方法中采用改进Bagging法并提出一种基于网络参数向量的相似度评价方法和选择性集成策略。最后通过UCI数据测试表明,同Bagging法和传统的全集成法相比,该方法拥有更为优秀的分类性能。

As its advantage, the training speed of extreme learning machine （ELM） is extremely fast. But sometimes its stability and precision can＇ t meet the requirement of practical application. In order to solve the problem, this paper introduces a solution for ELM when to be used in classification, in it the output weight matrix is improved with the evaluation factor of information in training results. Meanwhile, the hidden layer output matrixes competitive mechanism is added to improve the stability of ELM. For the sake of further improving ELM＇ s accuracy rate in classification, we propose a kind of selective ensemble extreme learning machine classifier by learning from the theory of neural network ensemble. In ensemble method, we adopt the improved Bagging and propose a subnet＇ s parameter vector-based similarity evaluation method and selective ensemble policy. Finally it is demonstrated by UCI data test that compared with Bagging and traditional all ensemble ELM, the solution proposed here has better performance in classification.