小时间序列的动态朴素贝叶斯分类器学习与优化-控制与决策.pdf

第32 卷 第1 期 控 制 与 决 策 Vol. 32 No. 1 2017 年 1 月 Control and Decision Jan. 2017 文章编号: 1001-0920 (2017) 01-0163-04 DOI: 10.13195/j.kzyjc.2015.1556 小时间序列的动态朴素贝叶斯分类器学习与优化 王双成 , 高 瑞, 杜瑞杰 (上海立信会计学院a. 数学与信息学院，b. 立信会计研究院，上海201620) 摘 要: 小时间序列在宏观经济领域普遍存在, 对小时间序列的分类预测也有着广泛的需求. 由于小时间序列蕴含 的信息不充分, 有效地提高小时间序列分类预测的可靠性非常困难, 目前也缺少这方面的研究. 针对这种情况, 在基 于引入平滑参数的高斯核函数估计属性边缘密度的基础上, 建立用于小时间序列分类预测的动态朴素贝叶斯分类 器, 并给出平滑参数的同步和异步优化方法. 实验结果表明, 优化能够显著提高小时间序列分类预测的准确性. 关键词: 贝叶斯网络；分类器；时间序列；高斯核函数；平滑参数 中图分类号: TP181 文献标志码: A Learning and optimization of dynamic naive Bayesian classifiers for small time series WANG Shuang-cheng , GAO Rui , DU Rui-jie (a. School of Mathematics Information，b. Lixin Accounting Research Institute ，Shanghai Lixin University of Commerce，Shanghai 201620，China) Abstract: The small time series exists generally in the field of macroeconomy. There are wide demands for the classification of small time series in macroeconomy. Because the information contained in the small time series is not sufficient, it is very difficult to effectively improve the reliability of small time series classification. In view of this situation, on the basis of using the Gaussian kernel function of introducing the smoothing parameter to estimate the attribute marginal density, the dynamic naive Bayesian classifier for small time series classification is presented, and the synchronous and asynchronous optimization method for smoothing parameters are given. The experimental results show that the classification accuracy of the small time series classifier can be improved significant