LINEX损失下一类分布族参数渐近最优的EB检验

目的 在LINEX损失函数下，讨论一类双边截断型分布族参数的经验Bayes(EB)检验问题。方法 构造适当的EB判决(检验)函数。结果 在经验Bayes检验问题中，将损失函数推广为LINEX损失，在适当的条件下证明了所构造的判决函数是渐近最优的。结论 LINEX损失函数具有比对称损失更广泛的意义，而且在一定条件下可以获得参数渐近最优的EB检验。     

Logistic回归模型的Bayes分析

本文应用Bayes观点讨论了logistic模型中的参数估计问题。     

Induction of hybrid decision tree based on post-discretization strategy

By redefining test selection measure, we propose in this paper a new algorithm, Flexible NBTree, which induces a hybrid of decision tree and Naive Bayes. Flexible NBTree mitigates the negative effect of information loss on test selection by applying postdiscretization strategy: at each internal node in the tree, we first select the test which is the most useful for improving classification accuracy, then apply discretization of continuous tests. The finial decision tree nodes contain univariate splits as regular decision trees, but the leaves contain Naive Bayesian classifiers. To evaluate the performance of Flexible NBTree, we compare it with NBTree and C4.5, both applying pre-discretization of continuous attributes. Experimental results on a variety of natural domains indicate that the classification accuracy of Flexible NBTree is substantially improved.     

基于Bayes网络的项目风险评估方法研究

本文研究了项目风险评估的建模方法 ,通过与其它传统的项目风险评估方法比较 ,针对项目风险评估的信息不完备、具有一定不确定性的特点 ,选择Bayes网络建立项目风险评估模型。本文内容适合于在不确定环境下 ,信息不完备时的项目风险评估     

非线性时间序列建模的混合GARCH方法

在文献[1]的基础上，首次提出混合广义自回归务件异方差（Mixture Generalized Autoregressive Conditional Heteroscedastic Model简记MGARCH）模型；给出并证明了MGARCH模型的一阶平稳性的充分必要条件及二阶平稳性的充分务件；给出该模型参数估计的EM算法：利用BIC定阶准则对MGARCH模型的各成份进行定阶；计算结果表明该模型对金融非线性时间序列中存在的变异率现象具有较强的描述能力，有广阔的应用前景。     

Privacy Quantification Model Based on the Bayes Conditional Risk in Location-Based Services

The widespread use of Location-Based Services(LBSs),which allows untrusted service providers to collect large quantities of information regarding users’locations,has raised serious privacy concerns.In response to these issues,a variety of LBS Privacy Protection Mechanisms(LPPMs)have been recently proposed.However,evaluating these LPPMs remains problematic because of the absence of a generic adversarial model for most existing privacy metrics.In particular,the relationships between these metrics have not been examined in depth under a common adversarial model,leading to a possible selection of the inappropriate metric,which runs the risk of wrongly evaluating LPPMs.In this paper,we address these issues by proposing a privacy quantification model,which is based on Bayes conditional privacy,to specify a general adversarial model.This model employs a general definition of conditional privacy regarding the adversary’s estimation error to compare the different LBS privacy metrics.Moreover,we present a theoretical analysis for specifying how to connect our metric with other popular LBS privacy metrics.We show that our privacy quantification model permits interpretation and comparison of various popular LBS privacy metrics under a common perspective.Our results contribute to a better understanding of how privacy properties can be measured,as well as to the better selection of the most appropriate metric for any given LBS application.     

Signal Diffusion Mapping: Optimal Forecasting with Time‐Varying Lags

We introduce a new methodology for forecasting, which we call signal diffusion mapping. Our approach accommodates features of real‐world financial data which have been ignored historically in existing forecasting methodologies. Our method builds upon well‐established and accepted methods from other areas of statistical analysis. We develop and adapt those models for use in forecasting. We also present tests of our model on data in which we demonstrate the efficacy of our approach. Copyright © 2015 John Wiley & Sons, Ltd.     

产生式与判别式分类方法比较研究

分类是数据挖掘领域研究的热点,产生式与判别式是数据挖掘中两种不同的分类模型。产生式模型具有通用性、灵活性及清晰的分层结构,学习得到的模型很容易满足模型解释要求;判别式模型没有明显的对系统中变量的基本分布建模的企图,仅仅对输入到输出之间映射的最优化感兴趣,可以提供更好的分类性能。从准确率、建模时间及渐进误差等方面对产生式与判别式分类方法进行了分析与比较,为研究人员在分类模型的选择上提供了参考。     

泊松分布下信息安全事件概率计算模型

从信息安全事件的概率分布规律出发,根据泊松分布的基本特征,通过数学证明了信息安全事件发生频数服从泊松分布,并采用国家互联网应急中心(CNCERT/CC)统计数据验证了这一理论结果.在此基础上,基于贝叶斯定理,建立了泊松分布下的信息安全事件概率计算模型.根据泊松分布的概率质量函数,计算了信息安全事件发生频数的先验概率分布;通过构建似然函数调整先验概率分布,得到信息安全事件发生频数后验概率分布;最后,采用CNCERT/CC统计数据验证了该模型的可行性和有效性.     

一种基于属性相关性度量的朴素贝叶斯分类模型

本文通过分析属性相关性的度量和属性约简,提出一种基于属性相关性度量的朴素贝叶斯分类模型EANBC。实验结果表明,与朴素贝叶斯分类模型相比,EANBC分类模型具有较高的分类正确率。