This book dilates on its underlying research which aimed to innovate a new general methodology for deriving generalised software metrics models from past empirical software metrics data. These generalised software metrics models are to predict the target software metric(s) of anyJuture software project from the project's predictor software metric(s) always in a best-effort, best-accuracy and best-consistency manner whether all, only some or even none of these required predictor software metric(s) is~are available for that future project and or projects in the past empirical software metrics data. A software project's predictor software metric(s) indicate(s) or measure(s) the project's relevant software engineering factor(s).

Foreword

Table of Contents

List of Illustrations

List of Tables

Acknowledgements

Abbreviations

1 Introduction

 1.1 Software Metrics

 1.2 Software Metrics Models

 1.3 Importance of Software Metrics Models

 1.4 Existing Software Metrics Models

 1.4.1 Function Point Analysis

 1.4.2 Inferential Statistics

 1.4.3 Neural Networks

 1.4.4 Fuzzy Logic Systems

 1.4.5 Hybrid Neuro-fuzzy Systems

 1.4.6 Rule-based Systems

 1.4.7 Case-based Reasoning

 1.4.8 Regression and Classification Trees

 1.5 Generalised versus Existing Software Metrics Models

 1.5.1 For Developers of Software Metrics Models

 1.5.2 For End-users of Software Metrics Models

 1.6 Overview of the Approach of this Research

2 Terminologies and Typography

 2.1 Terminologies

 2.2 Typography

 3 The Objectives of the Research

4 Summary of the General Methodology Innovated in this Research

5 Theories Underlying the General Methodology Innovated in this Research

 5.1 The "EM Algorithm for the General Location Model".

 5.1.1 Theory

 5.1.2 Application to this General Methodology

 5.1.3 What Ifa Future Project Has No Missing Values

 5.2 Transformation on the Continuous Software Metrics and Testing Multivariate Normality.

 5.2.1 Power Transformations on Individual Continuous Software Metrics

 5.2.2 Testing Univariate Normality of the Individual Continuous Variables

 5.2.3 Testing Multivariate Normality of the Continuous Variables

 5.3 Detection and Elimination of the Multivariate Outlier(s) in Respect of the Continuous Variables

 5.3.1 Theory

 5.3.2 Application to this General Methodology

 5.4 Linear LS Regression of Each Continuous Independent Variable on All Other Continuous Independent Variable(s)

 5.4.1 Theory

 5.4.2 Application to this General Methodology

 5.5 Coefficient of Determination R2 for the Linear LS Regression of the Dependent Variable on the Continuous Independent Variable(s)

 5.5.1 Theory

 5.5.2 Application to this General Methodology

 5.6 Data Splitting, Mean Magnitude of Relative Error and Pred

 5.6. l Theory

 5.6.2 Application to this General Methodology

 5.7 The Bootstrap Method

 5.7.1 Bootstrap Procedures

 5.7.2 Bootstrap Analysis

 5.7.3 Confidence Intervals

 5.7.4 Test of Hypotheses

 5.8 Plots of the Residuals versus the Predicted Dependent Variable

6 Findings and Results

 6. l Stage 1 : Data Sourcing

 6.1.1 Data Sources

 6.1.2 ISBSG

 6.1.3 Content of the ISBSG 6 Data

 6.2 Stage 2: Rectification of Software Metrics Data

 6.2.1 Shortlisting Software Metrics

 6.2.2 "Filtering" Software Projects

 6.2.3 Transformation on the Continuous Software Metrics and Testing Multivariate Normality

 6.2.4 Detection and Elimination of the Multivariate Outlier(s) in Respect of the Continuous Variables

 6.3 Stage 3: Constructing the Candidate Models

 6.3.1 For the Intended Model with the PDR as the Target Metric

 6.3.2 For the Intended Model with the "Summary Work Effort" as the Target Metric

 6.4 Stage 4: Selecting and Optimising Candidate Models

 6.4.1 For the Intended Model with the PDR as the Target Metric 1

 6.4.2 For the Intended Model with the "Summary Work Effort" as the Target Metric

 6.5 Stage 5: Analysis of Software Engineering Factors

 6.5.1 For the Intended Model with the PDR as the Target Metric

 6.5.2 For the Intended Model with the "Summary Work Effort" as the Target Metric

7 Discussion on the Findings and Results

 7.1 Limitations of the Findings and Results

 7.1.1 Unavoidably Biased Sampling

 7.1.2 Unavailability of"Ideal" Software Metrics

 7.1.3 Evolving Software Engineering/Development Technologies, Tools and Equipment

 7.1.4 Extrapolation

 7.1.5 No Causality Relationship Established

 7.2 Comparision between the Existing Software Metrics Models and Generalised Models

 7.2.1 Empirical Prediction Accuracy and Consistency of the Existing Software Metrics Models

 7.2.2 Empirical Prediction Accuracy and Consistency of the Generalised Models

 7.2.3 Summarising the Comparison

 7.3 Foreseeable Improvement Areas for the GeneralisedModels of this Research

 7.3.1 Complexity Measurement

 7.3.2 Number of Bootstrap Samples

 7.4 Other Comments on the Generalised Models

8 Conclusion

Appendix A The Sweep Operator

Appendix B Listing of the Scripts Implemented in this Research

B.1 The Script "MultiVarNorm.SBS" to Test Multivariate Normality of the Continuous Variables

B.2 The Script "LMSSwMD.SBS" to Detect and Eliminate the Multivariate Outlier(s)

B.3 The Script "EMContCatSwMD2.SBS" of Functions Subsidiary to the Scripts "EMContCatSwMDAcc.SBS" and "EMCont CatSw MDConflnt. SBS".

B.4 The Script "EMContCatSwMD3.SBS" of Functions Subsidiary to "EMContCatSwMD2.SBS," "EMContCatSwMDAcc.SBS" and "LSLMSSwMDAcc.SBS".

B.5 The Script "EMContCatSwMDAcc.SBS" to Construct the Candidate and "Prospective" Optirnised Models through the "EM Algorithm for the General Location Model".

B.6 The Script "EMContCatSwMDConflnt.SBS" to Optimise the "Winning" Candidate Models

B.7 The Script "LSLMSSwMDAcc.SBS" to Construct Software Metrics Models through the Linear LS and Linear LMS Regressions

References