版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
1、a,1,面向缺陷的软件系统可靠性管理规范的研究,寇纲 电子科技大学经济与管理学院,a,2,a,3,基于数据挖掘和多目标决策的软件风险评估和管理,Risk Assessment,Risk Management,What can be done and what options are available? What are the associated trade-offs in terms of all costs, benefits, and risks? What are the impacts of current management decisions on future optio
2、ns,What can go wrong? What is the likelihood that it could go wrong? What are the consequences? What is the time domain,Kaplan and Garrick 1981,Haimes 1991,Risk Communication (Data Mining,Risk Communication (MCDM,Yacov Haimes 2009,a,4,No Free Lunch (NFL) theorem,if algorithm A outperforms algorithm
3、B on some cost functions, then loosely speaking there must exist exactly as many other functions where B outperforms A” (Wolpert and Macready, 1995). In other words, there exists no single classifier that could achieve the best performance for all measures,a,5,Approach 1overview,Aim: Design a perfor
4、mance metric that combines various measures to evaluate the quality of classifiers for software defect prediction; Data: 11 datasets from NASA MDP repository; Tool: WEKA Techniques: Statistic,a,6,Approach 1Classifiers,Trees: classification and regression tree (CART), Nave Bayes tree, and C4.5 Functi
5、ons: linear logistic regression, radial basis function (RBF) network, sequential minimal optimization (SMO), Support Vector Machine (SVM), and Neural Networks Bayesian classifiers: Bayesian network and Nave Bayes lazy classifiers: K-nearest-neighbor Rules: decision table and Repeated Incremental Pru
6、ning to Produce Error Reduction (RIPPER) rule induction,a,7,Approach 1Step 1,For a specific dataset i (i=1,2,11), and a specific performance measure j (j=1,2,13), Do t test for pairs of classifiers (k=1, 2,13): (the statistical significance is set as 0.05,I f C_1 performs better at measure j than C_
7、2,The top three ranking classifiers are assigned to the score of 3, 2, and 1, respectively,a,8,Approach 1Step 2,For a specific dataset i,The larger the “Sum_rank”, the better the classifier is. The value of “Sum_rank” is normalized,Sum,a,9,Approach 1Step 3,For a specific dataset i,Sum,The lager the
8、score, the better the classifier,a,10,Approach 1Results,a,11,Approach 1conclusion,The best result for a given dataset according to a given measure may perform poorly on a different measure. Neural network and SVM have longer training time than other classifiers in general. No classifier yielding the
9、 best measures across the 11 datasets. SVM (functions.LibSVM), K-nearest-neighbor (lazy.IBk), and C4.5 (trees.J48) ranked the top three classifiers based on the experiment,a,12,Approach 2why,Experimental results have shown that ensemble of classifiers are often more accurate and robust to the effect
10、s of noisy data, and achieve lower average error rate than any of the constituent classifiers. However, inconsistencies exist in different studies and the performances of learning algorithms may vary using different performance measures and under different circumstances,a,13,Approach 2Overview,Aim:
11、Evaluate the performance of ensemble classifiers for software defect detection; Data: 11 datasets from NASA MDP repository; Tool: WEKA, Matlab 7.0 Techniques: MCDM Tool: AHP,a,14,Approach 2ensemble methods,Bagging It combines multiple algorithms by taking a plurality vote to get an aggregated single
12、 predictor. randomly sampling. Boosting In boosting, however, weights of training instances change in each iteration to force learning algorithms to put more emphasis on instances that were predicted incorrectly previously and less emphasis on instances that were predicted correctly previously Stack
13、ing minimizing the generalization error rate of one or more algorithms; Can different types of learning algorithms; Vote,a,15,Approach 2AHP,The analytic hierarchy process (AHP) is a multi-criteria decision making approach that helps decision makers structure a decision problem based on pairwise comp
14、arisons and experts judgments,a,16,Approach 2Pairwise comparisons of performance measures,a,17,Approach 2Priorities of AdaBoost classifiers (Group 1,a,18,Approach 2 Priorities of bagging classifiers (Group 2,a,19,Approach 2 Priorities of stacking, voting and individual classifiers (Group 3,a,20,Appr
15、oach 2 Priorities of classifiers of the top 5 classifiers from each group,a,21,Approach 2 Conclusions,Ensemble methods can improve the classification results for software defect prediction in general; AdaBoost ensemble method gives the best results. Tree and rule based classifiers perform better tha
16、n other types of classifiers in the experiment. Stacking and voting can improve classification results and provide relatively stable outcomes, but the results are not as good as AdaBoost and bagging. The ranking conducted by the AHP may change in different settings of comparisons. When the set of al
17、ternative classifiers change, the relative ranking of algorithms may change, especially when the difference between two classifiers is statistically significant,a,22,Approach 3why,The ranking conducted by the AHP may change in different settings of comparisons. There are lots of MCDM methods availab
18、le to evaluate the classification results. We want to find out whether they make the same choice,a,23,Approach 3MCDM Methods,DEA Data envelopment analysis They evaluate the efficiency of decision making units (DMUs) through identifying the efficiency frontier and comparing each DMU with the frontier
19、; CCR PROMETHEE II Preference Ranking Organisation METHod for Enrichment of Evaluations,a,24,Approach 3weights,Weights reflect the preference of decision makers,a,25,Approach 3Results of the DEA technique,a,26,Approach 3Results of the TOPSIS technique,a,27,Approach 3Results of the ELECTRE I techniqu
20、e,a,28,Approach 3Results of the PROMETHEE II technique,a,29,Approach 3 Conclusion,The boosting of CART and the boosting of C4.5 decision tree are ranked as the most appropriate algorithms for software defect datasets. The four MCDM methods generate similar top-ranked classification algorithms while
21、produce different ranking for some classifiers for the selected software defect datasets. TOPSIS and PROMETHEE II may be more appropriate than DEA and ELECTRE I for the given task since they provide a complete ranking of algorithms,a,30,Approach 4 FAMCDM: A Fusion Approach of MCDM Methods to Rank Mu
22、lticlass Classification Algorithms,Since multiclass algorithms selection normally involves more than one criterion, such as accuracy and computation time, the selection process can be modeled as a multiple criteria decision making (MCDM) problem. While the evaluations of algorithms provided by vario
23、us MCDM methods are in agreement most of the time, there are situations where different MCDM methods generate very different results. To resolve this disagreement and help the decision maker to make the choices, this paper proposes a fusion approach to produce a weighted compatible ranking of multic
24、lass classification algorithms. A large-scale multiclass network intrusion prediction task is used as an illustrative case. The results of the experimental study suggest that MCDM methods are useful tools for evaluating multiclass classification algorithms and the fusion approach proposed is capable
25、 to identify a compromised solution in conflicting rankings generated by different MCDM methods,a,31,Approach 4 FAMCDM: A Fusion Approach of MCDM Methods to Rank Multiclass Classification Algorithms,In the first step, a selection of MCDM methods is applied to rank multiclass classification algorithm
26、s. If there are strong disagreements among MCDM methods, the different ranking scores generated by MCDM methods are used as inputs for the second step. The goal of the second step is to determine the weights of different MCDM methods. This paper utilizes the combinational evaluation model to find the weights for each MCDM method. The third step of the fusion approach uses the weights obtained from the second step to get secondary rankings of algorithms. Rankin
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 4000球罐施工方案(3篇)
- 促销外套活动策划方案(3篇)
- 架空層施工方案(3篇)
- 医学微生物学PBL实验教学中的探究共同体
- 茶艺旅游管理就业方向
- 乘风破浪光芒万丈-三八女神节励志主题演讲课件
- 护理院感应知应会考试试题(含答案解析)
- 《运输发展》-学习情境4
- 干部纪律教育主题活动纪实
- 安全培训机构教务管理
- 收心归位聚合力 实干奋进创未来总经理在2026年春节复工全体员工大会上的致辞
- 2025-2026学年北京市通州区高三(上)期末语文试卷
- Unit 1 Time to relax(复习课件)英语新教材人教版八年级下册
- 低压电工证理论试题库及答案
- 起重司索信号工培训课件
- 肥胖患者麻醉管理专家共识(2025版中国麻醉学指南与专家共识)
- 2026年湖南城建职业技术学院单招职业技能考试备考试题带答案解析
- 焦化厂电工培训课件教学
- 涉密文件销毁设备选型与管理
- 古代汉语(全套课件220P)
- 部编版初中语文文言文句子翻译复习公开课课件
评论
0/150
提交评论