基于成本代价的分布式逻辑回归算法在煤炭系统中的研究与应用.pdfVIP

第 32 卷第 8 期 煤炭技术 Vo1.32,No.08 2013 年8 期 Coal Technology A略ust，20 13 基于成本代份的分布扎逻辑回归算锋在 s煤炭系统中 的研究与应用 王浩亮，王丽莉 (东莞理工学院城市学院，广东东莞 523脱)()) 摘 要:随着互联网络的发展，煤炭系统也开发了多种网络销售系统，并且数据规模呈现指数级上升趋势。逻辍回归 算法是被广泛应用的分类算法之一。但是，传统的逻辑回归算法没有考虑预测错误代价，将不同的预测视为同等戚本 代价，会一定程度上影响算法的预测精度。文章提出了基于成本代价的逻辑回归算法，诙算法考虑了预测错误代价完 成煤炭数据集重新标签并完成预测工作。并且，针对海量规模煤炭数据，提出了基于云计算平台的分布式算法。通过 大量的实验结果，表明提出的算法能够明显降低预测算法的代价成本，并且没有影响算法的预测精度。 关键词:逻辑回归算法;戚本代价;分类;云计算; MapReduce 申图分类号:TP3 1 1.1 3 文献标识码 :A 文章编号 :1 鹏-8725(2013)08-0205 -03 Research and Application of Cost ?ased Distributed Logistic Regression AIgorithm in Coal System WANG Hao-lia吨， WANG Li-li (City College of Dongguan University of Technology, Dongguan 523创泪. China) Abstract:With the development of internet, coal systems also develop many web sale system.The data is increasing exponentially. Logistic regression is one of the popular classification algorithms. However, traditional logistic regression algorithm does not consider .the cost of wrong prediction, and they think different predictions have same cost, which influence the prediction accuracy. In this pape瓦 we propose a cost based logistic regression algorithm, and it considers the cost of wrong prediction and relabel 出e training dataset to complete the prediction work. Besides, focusing on the huge scale coal data, we also propose distributed algorithm based on cloud computing platform. Experimental results show that our algorithm could low the cost of prediction and it does not influence the prediction accuracy. Key words:Logistic regression algor