用于軟件測(cè)試的資源通常是有限的,但是軟件測(cè)試往往需要消耗大量費(fèi)時(shí)昂貴的軟件模塊。此外,由于軟件開發(fā)過程中測(cè)試往往并不充分,導(dǎo)致傳統(tǒng)的軟件測(cè)試手段并不足以保證軟件的質(zhì)量。因此,早期軟件產(chǎn)業(yè)發(fā)展階段中的軟件故障自動(dòng)預(yù)測(cè)技術(shù)在當(dāng)前仍然存在�，F(xiàn)今軟件故障預(yù)測(cè)主要用于設(shè)定與優(yōu)化軟件測(cè)試的優(yōu)先級(jí),以充分利用有限的測(cè)試資源并盡可能地提升軟件質(zhì)量。在這方面,機(jī)器學(xué)習(xí)方法得到了較為廣泛的應(yīng)用。然而,將機(jī)器學(xué)習(xí)方法應(yīng)用于精確的軟件故障預(yù)測(cè)對(duì)數(shù)據(jù)質(zhì)量有較高的要求。遺憾的是,真實(shí)的數(shù)據(jù)集卻質(zhì)量欠佳。在軟件故障預(yù)測(cè)中,人們可以借助已標(biāo)注的實(shí)例來構(gòu)建一個(gè)模型以預(yù)測(cè)迄今尚未發(fā)現(xiàn)的新實(shí)例的類別。如果用于訓(xùn)練預(yù)測(cè)模型的數(shù)據(jù)集受到污染,則會(huì)給訓(xùn)練階段和最終得到的模型都帶來不利影響。一個(gè)可預(yù)期的結(jié)果是最終得到的模型精度必然不高。因此,提升數(shù)據(jù)集質(zhì)量的一個(gè)有效策略是對(duì)帶有缺陷數(shù)據(jù)的數(shù)據(jù)集進(jìn)行清洗,主要是通過偵測(cè)數(shù)據(jù)集中可能存在的問題并消除這些問題來實(shí)現(xiàn)的。通過對(duì)現(xiàn)有軟件故障預(yù)測(cè)領(lǐng)域相關(guān)文獻(xiàn)的綜述我們發(fā)現(xiàn),分類在此領(lǐng)域中的大多數(shù)場(chǎng)合有著不可替代的重要作用。在一些特殊的場(chǎng)合,一些輔助的策略在應(yīng)對(duì)數(shù)據(jù)質(zhì)量挑戰(zhàn)中也不可或缺。一些無足... 

【文章來源】：西南交通大學(xué)四川省 211工程院校 教育部直屬院校

【文章頁數(shù)】：126 頁

【學(xué)位級(jí)別】：博士

【文章目錄】：
摘要
Abstract
List of Abbreviations
List of Symbols
1 Introduction
    1.1 Background
    1.2 Data Quality and Software Fault Prediction
    1.3 Motivation
    1.4 Dissertation Objectives
    1.5 Research Significance
    1.6 Dissertation Outline
2 Related Work
    2.1 Software Testing.
    2.2 Software Testing Goal
    2.3 Software Fault Prediction
        2.3.1 Common Software Fault Prediction Process
        2.3.2 Machine Learning Application in Software Fault Prediction
        2.3.3 Software Metrics
    2.4 Data Quality Challenges
        2.4.1 High Dimensionality
        2.4.2 Class Imbalance Problem.
        2.4.3 Noise Filtering
        2.4.4 Instance Selection
        2.4.5 Outlier Analysis
    2.5 Model Validation Techniques
    2.6 Performance Evaluation Metrics
    2.7 Summary
3 A Combined-Learning Based Framework for Improved Software Fault Prediction
    3.1 Overview
    3.2 Hypothesis
    3.3 Combined-Learning Based Framework
        3.3.1 Software Metrics
        3.3.2 Feature Selection Techniques
        3.3.3 Data Balancing
    3.4 Experimental Design
    3.5 Analysis and Discussions
        3.5.1 Classification Performance on mc1 SCM
        3.5.2 Classification Performance on jm1 SCM
        3.5.3 Classification Performance on camel-1.6 OOM
        3.5.4 Classification Performance on prop-4 OOM
        3.5.5 Classification Performance on ComML and ComLC Metrics
        3.5.6 Comparison:SCM and OOM
    3.6 Summary
4 A Three-Stage Based Ensemble Learning for Improved Software Fault Prediction
    4.1 Overview
    4.2 Three-Stage Based Ensemble Learning Framework
        4.2.1 Stage One:Information Gain Based Feature Filtering
        4.2.2 Stage Two:Synthetic Faulty Prone Over-sampling Based Data Sampling
        4.2.3 Stage Three:Fusion of Classifiers Strategy Based Noise Filtering
    4.3 Experimental Design
    4.4 Analysis and Discussions
        4.4.1 Performance in Stage One
        4.4.2 Performance in Stage Two
        4.4.3 Performance in Stage Three
        4.4.4 Multiple Comparison of Three-Stages Using Different Performance Met-rics
    4.5 Summary
5 Software Fault Prediction Using Hybrid Data Reduction Approaches
    5.1 Overview
    5.2 Hybrid Data Reduction Based Framework
        5.2.1 Instance Selection
        5.2.2 Outlier Analysis
    5.3 Experimental Design
    5.4 Analysis and Discussions
        5.4.1 Performance of Single Data Reduction Approach
        5.4.2 Performance of Two-Hybridized Data Reduction Approaches
        5.4.3 Performance of Three-Hybridized Data Reduction Approaches
        5.4.4 Multiple Comparison and Statistical Test of Eleven Data Reduction Ap-proaches
    5.5 Summary
6 Conclusions and Future Works
    6.1 Conclusions
    6.2 Future Works
Acknowledgements
References
List of Publications
Research Fundings



本文編號(hào)：3551891