信用评估中的鲁棒赋权自适应Lp最小二乘支持向量机方法

消费者信用评估是金融风险管理和信用产业竞争的一个重要方面.信用评估数据中常带有噪声点,并且其类别是不均衡的.最小二乘支持向量机是一个被广泛应用的分类模型,其模型简单,求解速度快,但鲁棒性差.本文提出了一个鲁棒赋权自适应L_p最小二乘支持向量机模型,能够适应信用评估样本数据库类别不均衡的特点,可以有效处理信用评估数据中带有噪声点的问题.在仿真数据和三个信用数据库上的实证分析表明,本文所提出的模型具有较好的鲁棒性和分类能力.     

客户信用评估半监督协同训练模型研究

在现实的很多信用评估问题中,由于对样本进行类别标记需要花费大量的人力、财力和物力,往往只能获取少量有类别标签的样本来训练分类模型,而把数据库中大量无类别标签的客户样本舍弃。为解决这一问题,本研究引入半监督学习技术,并将其与多分类器集成技术中的随机子空间方法(Random Subspace, RSS)相结合,构建了类别不平衡环境下基于RSS的半监督协同训练模型RSSCI。该模型主要包括三个阶段:1)使用RSS方法训练得到若干基本分类器;2)从大量无类别标签数据集中选择性标记一部分最合适的样本加入到原始训练集中;3)在最终的训练集上训练分类模型,并对测试集样本进行分类。在三个客户信用评估数据集上进行实证分析,结果表明,RSSCI模型的信用评估性能不仅优于常用的监督式集成信用评估模型,也优于已有的一些半监督协同训练信用评估模型。     

基于相似样本归并的大样本混合信用评估模型北大核心CSCDCSSCI

当前面向大样本设计的信用评估模型,大多没有深入探究大样本的分布特征,只是简单地将传统评估方法应用在大样本上.首先提出了用于描述大样本分布特征的相关属性集、边界向量等若干概念及定义,并证明了其主要性质.之后在两个大样本数据集的基础上,研究了样本在相似性方面的分布特征,最后设计了一种大样本混合信用评估模型——HLSCE模型.HLSCE模型认为在大样本数据集中,样本的同一属性在不同局部区域内,对分类性能的贡献是不同的.具体地,HLSCE模型根据各样本与边界向量的相似性差异,结合生物启发式算法,将样本归并划分为若干子集并分别在其上训练基分类器.实证研究表明,HLSCE模型的分类精度相比于现有的代表性信用评估模型更高,同时也具有更为优越的平衡性与稳定性.     

数据分类中的特征选择算法研究

本文应用半正定规划支持向量机模型,将核函数特征子空间的组合作为核映射矩阵,提出一种新的将特征选择整合在数据分类过程中的学习算法。首先,将样本按其特征进行分组,计算每组样本子集的核矩阵;然后将这些核矩阵线性组合后加入基于半正定规划的支持向量机模型中,利用半正定规划支持向量机学习器求解得到各子特征空间的权重系数,其次,根据特征权重系数建立特征贡献度和支持度用于特征选择并控制分类准确率、特征数量和对不同类别样本的分类能力;最后根据最优分类准确率、最少特征数量、最佳泛化能力三项不同目标计算所对应的特征数量和分类结果。实证中采用医学、植物学、文本识别和信用等领域数据以及人工数据集比较该方法和SFS、Relief-F以及SBS算法的特征选择效果。结果表明,在实际数据中,本文提出的方法不但能够保持较好的分类学习效果,而且可以比SFS、Relief-F以及SBS特征选择算法的特征子集数目大幅减少;在人工数据中,该方法可以正确地选出真正的特征,去除噪声特征。     

基于相似样本归并的大样本混合信用评估模型

当前面向大样本设计的信用评估模型,大多没有深入探究大样本的分布特征,只是简单地将传统评估方法应用在大样本上.首先提出了用于描述大样本分布特征的相关属性集、边界向量等若干概念及定义,并证明了其主要性质.之后在两个大样本数据集的基础上,研究了样本在相似性方面的分布特征,最后设计了一种大样本混合信用评估模型——HLSCE模型.HLSCE模型认为在大样本数据集中,样本的同一属性在不同局部区域内,对分类性能的贡献是不同的.具体地,HLSCE模型根据各样本与边界向量的相似性差异,结合生物启发式算法,将样本归并划分为若干子集并分别在其上训练基分类器.实证研究表明,HLSCE模型的分类精度相比于现有的代表性信用评估模型更高,同时也具有更为优越的平衡性与稳定性.     

仿EM的多变量缺失数据填补算法及其在信用评估中的应用

数据缺失会显著降低信用评估模型的准确性和可用性,尤其是多变量同时有数据缺失时。本文针对模型应用阶段的多变量数据缺失问题,提出了一种新的数据填补算法。该算法由两阶段构成：准备阶段和数据填补阶段。在准备阶段,算法基于朴素贝叶斯方法以初始数据集进行训练,对每个可能缺失的变量构建起相应的单变量预测估计模型;而数据填补阶段则借鉴了EM算法的思想,利用前期的单变量预测估计模型,对给定的多变量数据缺失样本进行交替迭代,逐步填补更新。理论证明,该算法具有单调收敛性。以人人贷数据集和UCI提供的德国和澳大利亚两个信用评估基准数据集为例,将其与众数填补法、EM填补法进行性能对比实验,结果表明本文方法的数据还原性能和填补后信用评估准确性都明显更优。这为解决信用评估时的数据多变量缺失问题提供了一种更好的处理方法。     

半模糊超球支持向量机多类分类方法研究

针对支持向量机在大类别模式分类中存在的问题,提出了一种基于半模糊核聚类的超球支持向量机分类方法.该方法首先利用半模糊核聚类方法对样本进行预处理,完成边缘样本的选取,进而以所选样本为训练样本进行超球支持向量机训练,从而有效提高分类器的性能.实验表明,该方法比标准支持向量机多类分类方法具有更高的速度和精度.     

面向巴塞尔新资本协议的自优化神经网络信用评估方法

巴塞尔新资本协议中对信用风险评估使用的数据作了明确规定,而我国银行业目前所积累的数据还不能达到协议的要求;立足于我国银行业的现状,在新协议框架范围内探索可行的信用评估方法.对原有神经网络算法加以改进,提出自优化神经网络方法,该方法能较好地适应时变数据和自动优化神经网络评估模型,用真实的客户信用数据试验也表明该方法比普通神经网络有较高的准确性.     

遗传繁衍样本策略及神经网络信用评价研究

提出遗传算法繁衍样本策略,克服了小样本建立模型的缺陷,建立适合于我国商业企业的信用评分指标体系,依据该指标体系和遗传优化样本建立了基于反向传播回归神经网络的信用评估模型,利用样本公司实际指标数据对该模型的评分效果进行了比较研究.     

基于最大指标区分度与最优相对隶属度的上市公司信用风险研究

信用风险又称违约风险，是评价一笔债务由于某些因素遭受损失的可能性，因此信用评价体系既要有违约鉴别能力，又能给出违约产生的原因。本文涉及的问题一是如何在众多指标中筛选出对违约状态鉴别最大的指标组合，二是如何对指标赋予权重，使得权重体现指标真实的重要程度，三是如何选取兼顾分类性能和可解释性的模型。本文的方法一是通过指标区分度最大的广义F-Score、指标子集间冗余度最小的平均相关系数遴选出对违约状态鉴别能力最大的指标体系；二是通过构建组合赋权模型，以指标信息含量最大、赋权结果一致性最大为目标函数，得到一组最优的指标权重；三是以客户与类别的广义加权距离最小为目标函数，求解非线性目标规划得到客户在不同类别中的相对隶属度，以此得到客户所属类别。上市公司样本实证分析表明：（1）从610个指标中筛选出的31个指标既具有违约状态的区分能力，又符合"信用5C"原则，是一个科学、合理的指标体系；（2）通过组合赋权得到31个指标的权重，企业内部财务因素对信用评价影响最大，"营运资本周转率"的权重值最大，是影响企业信用评价的关键指标，"监事会会议次数"、"派息比"、"审计意见类型"等指标都是影响信用评价的重要指标；（3）与逻辑回归、支持向量机等8种模型对比发现，本文提出的相对隶属度模型不仅具有良好的分类性能，而且是一个平衡分类性能、可解释性的模型。     

Criteria for selecting a production control philosophy

Logistic objectives constitute a compromise between having short leadtimes and a low volume of work in progress on the one hand and a high resource loading and due date obligation on the other. Miscellaneous production planning approaches offered today provide different combinations of logistic objectives for satisfying a manufacturing strategy. To combine the strength of the various approaches into a unified system, a new approach based on a dynamic and distributed production planning methodology is proposed. To customize the approach, various analyses and specifications have to be made and for that reason some important characteristics and criteria for analysing production control philosophies will be given.     

The social construction of the meaning of acute stressors: a qualitative study of the personal accounts of police officers using a stress counselling service

The police profession is one in which acute stressors are encountered more frequently than in other occupations. Using the personal accounts of 35 police officers attending an in-house stress counselling clinic, the aim of the present study was to provide a qualitative examination of how the institutional context of policing influenced the ways in which acute stressors signified to individual police officers experiencing felt distress. Using the framework of Rational Emotive Behaviour Therapy as an analytical tool, it is argued that beliefs contributing to the experience of felt distress are related to the way in which policing as both an identity and an activity is constructed through the police organizational culture. Not only do these constructions influence the ways in which officers perceive themselves and their environments, but they also operate at the collective level to 'normalize' some emotional responses and to 'pathologize' others which, it is argued, could impact upon the outcomes of interventions such as stress counselling.     

The moderating role of social support in Karasek's job strain model

This paper examines whether social support is a boundary-determining criterion in the job strain model of Karasek (1979). The particular focus is the extent to which different sources of social support, work overload and task control influence job satisfaction, depersonalization and supervisor assessments of work performance. Hypotheses are tested using prospective survey data from 80 clerical staff in a university setting. Results revealed 3-way interactions among levels of support (supervisor, co-worker, non-work), perceived task control and work overload on levels of work performance and employee adjustment (self-report). After controlling for levels of negative affect in all analyses, there was evidence that high levels of supervisor support mitigated against the negative effects of high strain jobs on levels of job satisfaction and reduced reported levels of depersonalization. Moreover, high levels of non-work support and co-worker support also mitigated against the negative effects of high strain jobs on levels of work performance. The results are discussed in terms of the importance of social support networks both at, and beyond, the work context.     

Job insecurity, coping resources and personality dispositions in occupational strain

This paper presents research on the relationships between the work-related stressor of perceived job insecurity and various indicators of occupational strain, taking into account employees' personality dispositions (trait negative and positive affectivity) and coping resources. Respondents were 222 Australian public servants surveyed during organizational restructuring that involved downsizing and threat to job certainty. The research was formulated within an adaptation of Osipow, Doty, and Spokane's (1985) framework of stress-strain-coping that included the possible direct as well as moderating effects of personality dispositions in reporting occupational strain (Parkes, 1990). Findings from hierarchical regression analyses indicate consistent significant independent effects of personality dispositions, coping resources and perceived job insecurity on various indicators of strain. There was also support for the moderating roles of negative affectivity and self-care in the relation between perceived job insecurity and physical strain. Implications for the role of dispositional factors, especially negative affectivity, and the utility of various coping resources in accounting for occupational strain in times of threatened job security are discussed.     

Introduction to Special Section on Sensitivity Analysis and Summary of NCSU/USDA Workshop on Sensitivity Analysis

This guest editorial is a summary of the NCSU/USDA Workshop on Sensitivity Analysis held June 11–12, 2001 at North Carolina State University and sponsored by the U.S. Department of Agriculture's Office of Risk Assessment and Cost Benefit Analysis. The objective of the workshop was to learn across disciplines in identifying, evaluating, and recommending sensitivity analysis methods and practices for application to food‐safety process risk models. The workshop included presentations regarding the Hazard Assessment and Critical Control Points (HACCP) framework used in food‐safety risk assessment, a survey of sensitivity analysis methods, invited white papers on sensitivity analysis, and invited case studies regarding risk assessment of microbial pathogens in food. Based on the sharing of interdisciplinary information represented by the presentations, the workshop participants, divided into breakout sessions, responded to three trigger questions: What are the key criteria for sensitivity analysis methods applied to food‐safety risk assessment? What sensitivity analysis methods are most promising for application to food safety and risk assessment? and What are the key needs for implementation and demonstration of such methods? The workshop produced agreement regarding key criteria for sensitivity analysis methods and the need to use two or more methods to try to obtain robust insights. Recommendations were made regarding a guideline document to assist practitioners in selecting, applying, interpreting, and reporting the results of sensitivity analysis.     

Job stress, job strain, and psychological withdrawal among Dutch university staff: towards a dualprocess model for the effects of occupational stress

This study examined the antecedents of job strain (emotional exhaustion, health complaints) and withdrawal behaviour (e.g. lowered organizational commitment) among a cross-sectional sample of 131 academic staff members of the law department of a large Dutch university. Conservation of resources theory (Hobfoll, 1989) provided the theoretical background for this study. Strains and withdrawal behaviours were expected to be most prominent among those who reported having few resources and/or who reported high job demands. Structural equation modelling revealed that this was indeed the case. As predicted, differential patterns of effects emerged for job demands and job resources. Analysis of the effects of four job-specific stressors revealed that especially the structural aspects of a staff member's teaching task (e.g. the number of students in their classes) contributed strongly to perceived job demands. Theoretical and practical implications of the study are discussed.     

Dermal Uptake of 18 Dilute Aqueous Chemicals: In Vivo Disappearance‐Method Measures Greatly Exceed In Vitro‐Based Predictions

Average rates of total dermal uptake (K_up) from short‐term (e.g., bathing) contact with dilute aqueous organic chemicals (DAOCs) are typically estimated from steady‐state in vitro diffusion‐cell measures of chemical permeability (K_p) through skin into receptor solution. Widely used (“PCR‐vitro”) methods estimate K_up by applying diffusion theory to increase K_p predictions made by a physico‐chemical regression (PCR) model that was fit to a large set of K_p measures. Here, K_up predictions for 18 DAOCs made by three PCR‐vitro models (EPA, NIOSH, and MH) were compared to previous in vivo measures obtained by methods unlikely to underestimate K_up. A new PCR model fit to all 18 measures is accurate to within approximately threefold (r = 0.91, p < 10^?5), but the PCR‐vitro predictions (r > 0.63) all tend to underestimate the K_up measures by mean factors (UF, and p value for testing UF = 1) of 10 (EPA, p < 10^?6), 11 (NIOSH, p < 10^?8), and 6.2 (MH, p = 0.018). For all three PCR‐vitro models, log(UF) correlates negatively with molecular weight (r² = 0.31 to 0.84, p = 0.017 to < 10^?6) but not with log(vapor pressure) as an additional predictor (p > 0.05), so vapor pressure appears not to explain the significant in vivo/PCR‐vitro discrepancy. Until this discrepancy is explained, careful in vivo measures of K_up should be obtained for more chemicals, the expanded in vivo database should be compared to in vitro‐based predictions, and in vivo data should be considered in assessing aqueous dermal exposure and its uncertainty.     

Systems Resilience for Multihazard Environments: Definition,Metrics, and Valuation for Decision Making

The United Nations Office for Disaster Risk Reduction reported that the 2011 natural disasters, including the earthquake and tsunami that struck Japan, resulted in $366 billion in direct damages and 29,782 fatalities worldwide. Storms and floods accounted for up to 70% of the 302 natural disasters worldwide in 2011, with earthquakes producing the greatest number of fatalities. Average annual losses in the United States amount to about $55 billion. Enhancing community and system resilience could lead to massive savings through risk reduction and expeditious recovery. The rational management of such reduction and recovery is facilitated by an appropriate definition of resilience and associated metrics. In this article, a resilience definition is provided that meets a set of requirements with clear relationships to the metrics of the relevant abstract notions of reliability and risk. Those metrics also meet logically consistent requirements drawn from measure theory, and provide a sound basis for the development of effective decision‐making tools for multihazard environments. Improving the resiliency of a system to meet target levels requires the examination of system enhancement alternatives in economic terms, within a decision‐making framework. Relevant decision analysis methods would typically require the examination of resilience based on its valuation by society at large. The article provides methods for valuation and benefit‐cost analysis based on concepts from risk analysis and management.     

Distinguishing traumatic, vicarious and routine operational stressor exposure and attendant adverse consequences in a sample of police officers

Problems in studying occupational stress within the police service are identified and the paucity of work on operational duties as potential stressors are discussed. The present study reports the results of a factor analysis of operational stressors (N = 601 serving British police officers) that revealed three factors: exposure to death and disaster; violence and injury; sexual crime. These were demonstrated to be reliable scales and were included in logistic regression models together with a range of demographic and psychological variables. Models were applied to men and women separately, which showed there to be different predictors of the likelihood of suffering distress (measured by the General Health Questionnaire, GHQ) in terms of the officer's gender and operational role. Overall the model for women officers was better at predicting psychological distress than that for men. These findings are related to aspects of the police occupational culture. Further discussion is offered that conceptualizes police operational stressors as traumatic, routine and vicarious. Finally, some implications are drawn for the provision of stress intervention in the light of this differentiation.     

UNIX系统性能监控简述及shell系统资源统计程序

本文简要阐述了unix系统中(主要以IBM的AIX为例)进行系统性能管理的几种命令行工具,并通过编写shell实例程序,运用其中提到的工具实现系统资源利用率的统计程序,为做好unix系统管理和调优提供参考.