STATE-OF-THE-ART IN EMPIRICAL VALIDATION OF SOFTWARE METRICS FOR FAULT PRONENESS PREDICTION: SYSTEMATIC REVIEW

International Journal of Computer Science & Engineering Survey (IJCSES) Vol.6, No.6, December 2015
DOI:10.5121/ijcses.2015.6601 1
STATE-OF-THE-ART IN EMPIRICAL VALIDATION OF
SOFTWARE METRICS FOR FAULT PRONENESS
PREDICTION: SYSTEMATIC REVIEW
Bassey Isong1
and Obeten Ekabua2
1
Department of Computer Sciences, North-West University, Mmabatho, South Africa
2
Department of Computer Science, Delta State University, Abraka, Nigeria
ABSTRACT
With the sharp rise in software dependability and failure cost, high quality has been in great demand.
However, guaranteeing high quality in software systems which have grown in size and complexity coupled
with the constraints imposed on their development has become increasingly difficult, time and resource
consuming activity. Consequently, it becomes inevitable to deliver software that have no serious faults. In
this case, object-oriented (OO) products being the de facto standard of software development with their
unique features could have some faults that are hard to find or pinpoint the impacts of changes. The earlier
faults are identified, found and fixed, the lesser the costs and the higher the quality. To assess product
quality, software metrics are used. Many OO metrics have been proposed and developed. Furthermore,
many empirical studies have validated metrics and class fault proneness (FP) relationship. The challenge is
which metrics are related to class FP and what activities are performed. Therefore, this study bring
together the state-of-the-art in fault prediction of FP that utilizes CK and size metrics. We conducted a
systematic literature review over relevant published empirical validation articles. The results obtained are
analysed and presented. It indicates that 29 relevant empirical studies exist and measures such as
complexity, coupling and size were found to be strongly related to FP.
KEYWORDS
Class, Empirical validation, Object-oriented metrics, Fault proneness.
1. INTRODUCTION
In today’s e-world, the importance of software technologies have been seen in different kinds of
productsand services used in everyday life. The exponential growth of software dependability
poses the demand for high quality from users and to meet this demand, today software has grown
in size and complexity [1][2][3][4]. This is because quality of software is the key determinant of
the success or failure of an organization [5]. However, guaranteeing high quality in this modern
age of large software systems development, increased difficulty, time and resource consumption
has become the order of the activity [4][5][6]. Given the size, the complexity and the constraints
imposed on the development, it is inevitable to deliver to customers software that have no faults
[1][3][4]. In particular, object-oriented (OO) products with its unique features could introduced
some faults that are hard if not impossible to find or pinpoint change impacts during maintenance.
Faults in software are errors introduced during the software development activity that can lead
software to fail or not meeting customers’ expectations. Though, it is difficult to find and fix
faults before product release, the earlier this is done the, lower the costs and the higher the
product quality would be [1][5][7][8][9][10][11] [12][13][14][15]. In software engineering, one
way to assure software quality cost-effectively is the use of software metrics.

2
Software metrics usage during development process, especially at the early phases is critical to
ensuring high quality in the software systems. There are used as a tool in software organizations
to assess software quality, monitor, control and take useful managerial and technical decisions
aimed at improving the software [16][17][18]. Existing software metrics are broadly classified
into traditional metrics and OO metrics [18]. Moreover, many OO metrics have been proposed
and developed for assessing OO design and codes quality [1][2][3][6][7][8][17][19][22]. OO
product metrics capture different software attributes such as class complexity, inheritance,
couplings and cohesions [10][16]. These structural properties are used to determine products
quality and complexity [10][19]. One of such OO metrics is the CK metric suit [16].
Albeit a greater amount of software faults found in software applications today are believed to
concentrate only on few classes of the system [10][20], what is more important is when such
faults are identified. In the world of OO systems, one viable approach used by engineers is to
identify faulty OO classes during the software development early stage through the construction
of quality models for prediction utilizing OO metrics and historical measures
[1][3][4][6][10][11][22][31]. The construction of these models can be used by organizations in
the identification of possible classes which are faulty either in the future applications or release
and to identify where resources are needed most [10]. Thus, it assist organizations to focus quality
improvement activities, make decisions, plan and schedule development activities in order to
produce high quality product within time and budget [10][19][21]. For instance, testing large
systems today is complex and time-consuming activity [5][6][10]. Therefore, predicting faulty
components early would allow organizations to take actions aim at mitigating against the high
risk posed by the faults which are likely to cause failure in the field. Such activities include
focusing testing and verification resources on such classes to avoid rework that could be costly
[10].
However, for OO design metrics to accurately predict faults in OO classes there have to be
empirically validated. That is, establishing which metrics are related to important external quality
attributes like class fault-proneness (FP). The essence is that, OO metrics are of no or little value
if such relationship is not empirically validated [10]. Nevertheless, few empirical validation
studies exist that have validated or re-validated OO metrics with respect to FP
[2][3][6][7][8][17][19][22][23][24]. In addition, these studies proposed and developed several
prediction models that make use of FP and OO metrics as dependent and independent variables
respectively. Among such validated OO metrics is the CK metric suite and size metric. Several
empirical studies in the literature has shown that some metrics are significantly or insignificantly
related to FP [2][3][6][7][8][17][19][22]. Furthermore, their findings appeared not to be
consistent [2]. For example, in one study a metric is considered related FP but insignificant
related to FP in another study. However, this could affect decision making in choosing directly
metrics that are associated with FP of a class. Hence, which of these metrics are actually related
to the FP of a class?
To establish OO design metrics that are related FP and are generic, this paper performed a
systematic literature review (SLR) using published empirical validation studies of CK +SLOC
metrics. The basis for this SLR is that the authors lack resources to perform empirical study on
real-world software systems, only few SLR on the CK + SLOC point of view exist within periods
considered and lastly, to bring together the state-of-the-art in fault prediction using FP and CK +
SLOC metrics. The study is specifically designed to assist software engineers take quick decision
regarding generic metrics that are suitable for fault prediction in a class when CK+SLOC metrics
are used.
The remaining part of the paper is organized as follows: Section 2 is the description of the metrics
used in this study, Section 3 is the research method used, Section 4 is analysis, Section 5 is the
study discussions and Section 6 is the conclusions.

3
2. METRICS STUDIED
The metrics considered in this study is the CK metric suit and the studies that have empirically
validated them. Moreover, the study also consider product size metric known as SLOC due to its
strong relationship with FP [1][17][22][25]. These metrics are shown on Table 1 alongside their
descriptions. They consist of six (6)OO design metrics and one size metric from the traditional
product metric.
Table 1. Metrics studied [16]
Metric Definition
CK:
Weighted Methods per Class (WMC)A count of methods implemented within a given class.
Coupling between Objects (CBO) CBO for a class is count of the number of other classes to which it is coupled and vice
versa.
Response for a Class (RFC) The count of methods implemented within a class plus the number of methods
accessible to an object class due to inheritance.
Lack of Cohesion (LCOM) For each data field in a class, the percentage of the methods in the class using that data
field; the %s are averaged then subtracted from 100 %.
Depth of Inheritance (DIT) The length of the longest path from a given class to the root in the inheritance
hierarchy
Number of Children (NOC) The NOC is the number of immediate subclasses of a class in a hierarchy.
Size:
Source Lines Of Code (SLOC) It counts the lines of code (nonblank and non-commented) in the body of a given class
and all its methods
3. RESEARCH METHODOLOGY
This study has been conducted by strictly following the guidelines for performing SLR offered by
Kitchenham et al [26][39]. SLR is a secondary study which provides the means to gather and
analyse a collection of published research findings that assist in answering stated research
questions. This SLR results will be useful in identifying the current state-of-the-art of the
empirical validation of the relationship between CK metrics, size measure and class FP. The steps
involve are discussed as follows.
3.1. Research Questions
This study is aim at providing empirical evidences from published studies in the literature to
identify which of the CK and SLOC metrics are strongly associated with class FP in terms of
significance level. Thus, the research questions intended to be answered are as follows:
RQ1: Which metric (s) within the CK metric suite and SLOC is related to the FP of a class?
This question is designed to provide answers on which metrics are significant or not significant
with FP of OO classes. This study will limit its findings to significance and insignificance
relationship regardless of if the relationship is positive, negative, weak, strong or severe.
RQ2: What techniques are being used to empirically validate the metrics in RQ1 and which is the
best?
This question will be used to explore the state-of-the-art in fault prediction using FP and CK and
SLOC with respect to the statistical techniques, models and variables used.
RQ3: To what extent have the metrics in RQ1 been validated?

4
This question is designed to elicit information about the state-of-the-art in fault prediction using
FP and CK and SLOC metrics with respect to the programming language used, the settings of the
validation, the type of systems used and the product release used.
RQ4: Of what relevance are the empirical validations of software metrics?
This question is designed to provide the relevance of empirically validating the relationship
between FP and CK and SLOC metrics
RQ5: Are there generic software metrics for predicting faulty classes?
This question is designed to provide answers on whether there exist validated OO metrics which
are generic in the prediction of FP of OO software systems. This is important to help developers
or managers make quick decisions during software development.
3.2. Search Strategy, Terms, Resources and Selection
Search strategy has the goal of ensuring that only relevant studies or articles appears in the search
results. In this study, we considered the review of 17-years’ efforts in empirical validation of CK
and SLOC metrics, between the period of January 1995 to December 2012. These periods were
strategically chosen with respect to the birth of CK metric suite and to sufficiently explore the
information provided within these periods. To this end, all studies published after the December,
2012 are not included. Another review will be carry out to cover the years after December 2012
in order to enable us perform comparisons on the state-of-the-art in fault predictions with those
periods.
However, search results are well documented to enhance the clarity of the search process and
avoid duplications. Search terms or strings were formulated and applied manually during the
process by following the steps suggested in [26]. For more details, refer to Isong and Ekabua [43].
Furthermore, we limited the search for relevant studies to electronic databases such as Google
Scholar, Compendex, Inspec and Scopus. There are subsets of databases largely recognized by
researchers worldwide and known to contain relevant journals and conferences articles within
computer science and software engineering. Databases such as IEEE Xplorer, Springer Link and
ACM were not searched directly since they were indexed or linked to the Engineering Village
database (Compendex and Inspec). Based on the study selection criteria designed, relevant studies
were selected during the review process to be used for data extraction. This is accomplished by
defining basic and detailed inclusion and exclusion criteria in accordance with the research
questions. In addition, quality assessment criteria was used to assess the quality of all included
studies. This is important to understanding the state of empirical validation of each included
study. In this case, each selected study is assessed against a number of checklist questions and
each question answered with Yes or No.
3.3. Data Extraction and Execution
This study designed data extraction form or template and used for information extraction. All
inconsistencies and difficulties encountered were resolved. Moreover, the extracted data was
checked at least twice by the authors. To achieve this, the authors independently carry out the
process involves in searching for articles that satisfied the defined inclusion and exclusion criteria
in the data extraction phase. The databases were scanned using the search terms/strings and the
basic defined inclusion and exclusion criterion on the articles to select relevant articles.
With the data extraction forms, each author performed validation on the extracted data in order to
accomplish inter-study consistency. All the information about the total number of results obtained
(selected and rejected articles) from each database were recorded in the search record. A total of
4683 articles that cited CK and SLOC metrics were retrieved after applying all search terms. At
first, studies were excluded after reading their title and abstracts. Furthermore, the remaining

5
studies were selected by applying thorough exclusion and inclusion criteria. More so, the
extracted data was compiled and organized quantitatively to answer the stated research questions.
Table 2 shows the list of selected Journals and Conferences papers considered in this SLR. Only
author’s first name was included due to space constraint. For more information, out of the 29
studies selected, 5 are from conference proceedings and 24 are from journals.
Table 2. Selected Articles
Id Ref. Year Author Title
1 [10] 2001 Emam et al The prediction of faulty classes using object-oriented design metrics
2 [19] 1998 Briand et al A Comprehensive Empirical Validation of Design Measures for OO Systems
3 [21] 2001 Emam et al The Confounding Effect of Class Size on the Validity of OO Metrics
4 [17] 2002 Yu et al Predicting FP using OO Metrics: An Industrial Case Study
5 [1] 2008 Zu et al An Empirical Validation of Object-Oriented Design Metrics for Fault Prediction
6 [22] 2000 Briand et al Exploring the relationships between design measures and software quality in OO systems
7 [6] 1996 Basili et al A Validation of Object-Oriented Design Metrics as Quality Indicators
8 [25] 2005
Gyimothy et al
Empirical Validation of OO Metrics on Open Source Software for Fault
Prediction
9 [23] 2007
Olague et al
Empirical Validation of Three Software Metrics Suites to Predict FP of OO Classes
Developed Using Highly Iterative or Agile Software Development Processes
10 [3] 2006
Zhou et al Empirical Analysis of Object-Oriented Design Metrics for Predicting High and Low
Severity Faults
11 [4] 2010 Singh et al Empirical validation of object-oriented metrics for predicting FP models
12 [8] 1999 Tang et al An Empirical Study on Object-Oriented Metrics
13 [5] 2003
Succi et al
Practical assessment of the models for identification of defect-prone classes in OO
commercial systems using design metrics
14 [2] 2003
Subramanyam et al Empirical Analysis of CK Metrics for OO Design Complexity: Implications for Software
Defects
15 [24] 2009
Aggarwal et al Empirical Analysis for Investigating the Effect of Object-Oriented Metrics on FP: A
Replicated Case Study
16 [27] 2001 Briand et al Replicated Case Studies for Investigating Quality Factors in OO Designs
17 [28] 2008
Olague et al An empirical validation of OO class complexity metrics and their ability to predict error-
prone classes in highly iterative, or agile, software: a case study
18 [11] 2010
Malhotra et al Empirical validation of OO metrics for predicting FP at different severity levels using
support vector machines
19 [29] 2012
S. Singh et al Validating the Effectiveness of OO Metrics over Multiple Releases for
Predicting FP
20 [12] 2008
Shatnawi et al The effectiveness of software metrics in identifying error-prone classes in post-release
software evolution process
21 [9] 2005
Janes et al Identification of defect-prone classes in telecommunication software systems using design
metrics
22 [30] 2009 English et al Fault Detection and Prediction in an Open-Source Software Project
23 [31] 2008 Goel et al Empirical Investigation of Metrics for Fault Prediction on OO Software
24 [32] 2011
Shaik et al Investigate the Result of Object Oriented Design Software Metrics on FP in Object Oriented
Systems: A Case Study
25 [33] 2011 Dallal, J.A Transitive-based object-oriented lack-of-cohesion metric
26 [34] 2010 Dallal et al An object-oriented high-level design-based class cohesion metric
27 [35] 2010
Zhou et al
On the ability of complexity metrics to predict fault-prone classes in Object-Oriented
systems
28 [36] 2007 Pai et al Empirical Analysis of Software Fault Content and FP Using Bayesian Methods
29 [37] 2012 Johari et al Validation of OO Metrics Using Open Source Software System: An Empirical Study

6
4. ANALYSIS AND RESULTS
This section presents the analysis of the findings in the SLR by answering the above stated
research questions.
4.1. CK and SLOC Metrics Relationship with Fault Proneness
RQ1: Which metric (s) within the CK metric suite and SLOC is related to the FP of a class? In
this study, 29 studies are considered on the basis of empirical validation of software metrics. In
these studies, 7 metrics (i.e. 6 CK metrics and 1 “traditional” metric) were empirically validated
as related the FP of OO class. However, the analysis carried out shows that some metrics are
significant, some strongly significant, some insignificant, while some are negatively significant
across the studies. Additionally, some studies categorized their findings in terms of significance
and insignificance based on the severity of the fault found such as high, medium, low and
ungraded [18][38]. But in this study there is no distinction as to whether a significance is positive
or negative and fault severity [3]. Nonetheless, analysis presented in Table 3, 4, 5, 6 and 7
indicates that some metrics are considered to be significant in some studies, insignificant in others
while few studies did not measured the metrics. The analysis of the finding is as follows:
Complexity measure: For WMC, the validation based on the hypothesis constructed confirms that
classes having more member functions or methods are more likely to have faults than classes with
small or no member functions. However, 22 studies confirmed WMC significance relationship
with the FP of OO classes, one study [2] found considered it to be insignificant while 6 others
studies did not consider it their studies.This is captured in Table 3.
Table 3. WMC Validation
Metric Significant Insignificant N/A
WMC [10],[40],[17],[1],[22],[6],[25],
[23],[3],[4],[8],[2],[24],[27],
[11],[29],[12],[31],[32],[35],[36],[37]
[2] [5],[9],[34],[19],[30],[33]
**N/A= not applicable
Coupling measures: Analysis indicates that 23 of the studies found CBO to be having strong
influence on class FP. The significance stems from the fact that a class that is highly coupled
tends to be more fault-prone than class that is loosely coupled. To this end, one study found CBO
to be insignificant but CBO was not measured in 5 studies.(See Table 4) Moreover, RFC was
found to be strongly significant related to class FP in 24 studies. The findings confirms that a
class with higher response sets tends to be more fault-prone than others with less response sets.
Interestingly, none of the studies found RFC insignificant but 5 of the studies did not measure
RFC.
Table 4. CBO and RFC Validation
CB0 [10],[19],[21],[17],[1],[22],[6],[25],[23],
[3],[4],[2],[24],[27],[11],[29],[12],[9],
[30],[31],[32],[36],[37]
[8] [5],[28],[35],[34],[33]
RFC [10],[19],[21],[17],[1],[22],[6],[11],
[23],[3],[4],[8],[5],[24],[27],[11],
[29],[12],[9],[30],[31],[32],[36],[37]
- [2],[28],[35],[34],[33]

7
Cohesion measure: Based on the analysis carried out in this study, it shows that 14 studies found
LCOM to besignificantly related to class FP. Nevertheless, only 4 studies considered LCOM to
be insignificant while 11 studies did not measure LCOM in their study. This is shown in Table 5.
The overall results confirmed that a class with low cohesion value is more likely to have faults
than class with high cohesion value.
Table 5. LCOM Validation
LCOM [19],[21],[17],[25],[23],[3],[4],[24],
[27],[11],[9],[32],[35],[34],[37]
[21],[6],[29],[31] [10],[1],[22],[8],[5],[2],
[28],[12],[30],[33],[36]
Inheritance measures: In the perspective of inheritance measure of a class, results has it that only
9 studies found DIT to be significantly (strong and weak) related to FP. However, about 15
studies considered it to be insignificant while 5 studies did not measure it. With emphasis on the
insignificance of DIT, it indicates that a class with higher number of inheritance hierarchy is not
likely to have faults. Furthermore, only 3 studies found NOC to be significantly related to FP
while 15 studies considered it insignificant. With the insignificance results, it show that a class
having a higher number of children is not likely to be fault-prone than others with less number of
children. The validation for both DIT and NOC are shown in Table 6
Table 6. DIT and NOC Validation
DIT [19],[22],[6],[25],[5],[2],[27],
[9],[37]
[10],[21],[17],[23],[3],[4],
[8],[24],[11],[29],[12],[30],
[31],[32],[36]
[1],[28],[35],[34],[33]
NOC [17],[22],[2] [19],[25],[23],[3],[4],[24],
[27],[11],[29],[12],[30],[31],
[32],[36],[37]
[10],[21],[1],[6],[5],[2],[28],[9],
[35],[34],[33]
Class Size measure: In this study, analysis indicates that SLOC of a class has a strong relationship
with FP and even more than OO metrics [1][17][22][25]. Consequently, about 17 studies
confirmed its significance on FP and no study considered it insignificant while 12 studies did not
measure it. (See Table 7) The implication of the results is that a class having a larger number of
lines of code is more likely to have faults than classes with small code lines.
Table 7. SLOC Validation
SLOC [17],[1],[22],[25],[3],[4],[2],[24],[27],
[28],[11],[29],[30],[31],[32],[33],[36]
- [10],[19],[21],[6],[23],[8],[5],[12],
[35],[34],[37]
4.2. Empirical Validation Techniques
RQ2: What techniques are being used to empirically validate the metrics in RQ1 and which is the
best? From the results of the analysis conducted, this study found that all the 29 studies selected
explicitly stated the techniques used in conducting their individual empirical validation. Table 8
shows the techniques used, metrics studied, the variables employed (dependent and independent)
and the tools employed for metric collection. However, different techniques were employed such

as machine learning, logistic regression (LR) and so on. Moreover, LR is the most reported
techniques used to construct pred
FP. With these findings, we can deduced that LR is the best and widely used statistical techniques
for predicting FP of a class CK+SLOC. Based on the analysis, about 76% of the studies used LR
model (i.e. univariate and multivariate), and other 24% is for other techniques. (
Fig
Also, the variables used in the models are the dependent and independent
explained in terms of cause and effect. In an experiment, an
input, while the dependent variable
independent variables are tested to valida
the prediction model. However, FP are used as the dependent variable in majority of the studies
(87%) and 10% used faults data (F), while only 3% used fault count (FC) a
variable.(See Fig. 2) For independent variable, CK and
specifically used.
Furthermore, metric collection method is considered to be critical to the accuracy of the metric
validated. From the analysis performed, it indicates that only
manually, 41% stated the tools used in the collection, while 5
metrics were collected. (See Fig. 3
Not_sta
ted
55%
OTHERS
24%
techniques used to construct predictive model that validate the relationship between
, we can deduced that LR is the best and widely used statistical techniques
l (i.e. univariate and multivariate), and other 24% is for other techniques. (See Fig
Figure 1. Statistical techniques used
Figure 2.Used dependent variables used
Also, the variables used in the models are the dependent and independent variables which can be
explained in terms of cause and effect. In an experiment, an independent variable is the cause or
dependent variable is the output or effect [39]. To this end, dependent and
independent variables are tested to validate if they are the actual effect and cause respectively by
(87%) and 10% used faults data (F), while only 3% used fault count (FC) a
g. 2) For independent variable, CK and SLOC metrics and others were
Furthermore, metric collection method is considered to be critical to the accuracy of the metric
validated. From the analysis performed, it indicates that only 4% of the studies collected metrics
manually, 41% stated the tools used in the collection, while 55% mentioned nothing about how
. 3).
Figure 3. Metric collection methods
FP
87%
F
10%
FC
3%
Tool
41%
Not_sta
ted
55%
Manual
4%
LR
76%
OTHERS
24%
8
e model that validate the relationship between metrics and
, we can deduced that LR is the best and widely used statistical techniques
Fig.1)
variables which can be
is the cause or
is the output or effect [39]. To this end, dependent and
te if they are the actual effect and cause respectively by
(87%) and 10% used faults data (F), while only 3% used fault count (FC) as dependent
SLOC metrics and others were
Furthermore, metric collection method is considered to be critical to the accuracy of the metrics
he studies collected metrics
% mentioned nothing about how

9
4.3. State of Metrics Validation
RQ3: To what extent have the metrics in RQ1 been validated? In this section, the state of the
metric validation are considered from different points of views: the study context, programming
language used, product release time and the study type. Table 8 present details of the metric
validation state.
4.3.1. Study Subjects and Context
In this study based on the analysis conducted, it shows that the empirical validation studies of CK
and SLOC metric’s relationship with FP have been carried out in both academia and non-
academia environments utilizing software products developed by either students or software
professionals respectively. The academic environment used mainly systems developed by
students while in the non-academia environment, either open source software (OSS) projects or
industrial software systems developed by professionals were utilized. In most of the selected
studies, product are either applications, components or middlewarethat ranges from OSS projects
like Mozilla [23][25][28], eclipse [12][30][33], NASA project [1][3][4][11][31][36] to
telecommunication systems [9][17][21]. Moreover, the systems have variable sizes ranging from
small to large sized systems.
Table 8. Validation details
Paper
Id Technique
Dependent
Variable
Independent
Variable
Metric Collection Tool Prog.
Language
Study
Type
Release
1 LR FP CK & Others (24 ) Java static analysis tool JAVA NR Pre
2 LR FP CK Metrics
M-System based on GEN++
C++
NR Pre
3 LR FP CK & Others commercial metrics collector C++ NR Pre
4 OLS,LDA FP CK OTHERS 8 Metric tool integrated with Rigi, JAVA NR Pre
5 OLS,ANFIS F CK SLOC - C++ NR Pre
6 LR FP CK & OTHERS 49 M-System based on GEN++ C++ NR Pre
7 LR FP CK M-System based on GEN++ C++ NR Pre
8 LR/ML FP CK SLOC Columbus C++ NR Pre
9 LR FP CK OTHERS Software System Markup Language JAVA NR Pre
10 LR/ML FP CK SLOC - C++ NR Pre
11 LR/ML FP CK SLOC - C++ NR Pre
12 LR FP CK-NOC - C++ NR Pre
13
PRM,
NBRM,ZINBRM F CK LOC
WebMetrics
C++
NR Pre
14 OLS F WMC,CBO,DIT,SIZE - C++, Java NR Pre
15 LR FP CK & OTHERS 49 Manual Java R Pre
16 LR FP CK & OTHERS 49
A tool based on the FAST parser
technology C++
R Pre
17 LR FP
WMC, LOC,
Complexity
Software System Markup Language
JAVA
NR Pre
18 ML (SVM) FP CK SLOC - C++ NR Pre
19 LR FP CK, LOC & OTHERS - C++ NR Pre
20 LR FP CK & OTHERS Borland Together JAVA NR Post
21
PRM,NBRM,
ZINBRM FP CK & Others
-
C++
NR Pre
22 LR FP CK & Others - java NR Pre

23 LR FP
24 LR FP CK, LOC & OTHERS
25 LR FP LCOM(CK) & others
26 LR FP LCOM(CK) & others
27 LR FP
28 PRM, NBRM FP,FC
29 LR FP
*LR: Logistic Regression; *ML: Machine Learning; *OLS, Ordinary Least Square; *LDA: Linear Discriminant Analysis;
*PRM: Poisson Regression Model; *NBRM: Negative Binominal Regression Model; *ZINBRM: Zeros
Binominal Regression Model; *ANFIS: Adaptive Neuro
R: Replicated, NR: Non-replicated; Pre: Pre-release; Post: Post
As presented in Table 8, further
79% in the non-academic environment
environment like the University of Maryland (UMLD) [3][19][22] and University School of
Information Technology (USIT) [24] and others (See Fig
used were written by students and 79% by mainly software professional few studies were
replicated studies that utilizes same data sets from public repository such as eclipse, NASA, and
others. (See Fig. 4). Also, the analysis shows that, 20% of the
projects, 33% are OSS and 47% are non
4.3.2. Programming Language Used
From the analysis carried out in this study, it indicates that only applications written with
programming languages: Java and C++ were
between OO design metrics and FP. To this end, it shows that the two O
Professi
onals(N
on-
Acade
Non-OSS
47%
CK & Others - C++
CK, LOC & OTHERS - java
LCOM(CK) & others - java
LCOM(CK) & others
-
java
WMC, LOC,
Complexity
-
JAVA
CK, SLOC - -
CK - JAVA
*LR: Logistic Regression; *ML: Machine Learning; *OLS, Ordinary Least Square; *LDA: Linear Discriminant Analysis;
*PRM: Poisson Regression Model; *NBRM: Negative Binominal Regression Model; *ZINBRM: Zeros-Inflated Negative
NFIS: Adaptive Neuro-Fuzzy Inference System *SVM: Support Vector Machine.
release; Post: Post-release.
, further analysis has shown that empirical validations is high
environment than only 21% of the validation occurred in the academia
University of Maryland (UMLD) [3][19][22] and University School of
Information Technology (USIT) [24] and others (See Fig. 4). Furthermore, 21% of the sy
that utilizes same data sets from public repository such as eclipse, NASA, and
). Also, the analysis shows that, 20% of the projects studied are student’s
projects, 33% are OSS and 47% are non-OSS systems.(See Fig. 5)
Figure 4. Validation environments
Figure 5. Projects validated
4.3.2. Programming Language Used
programming languages: Java and C++ were majorly used in the validation of the relationship
between OO design metrics and FP. To this end, it shows that the two OO languages have
Student
s(Acade
mia)
21%
Professi
onals(N
Acade…
Students
20%
OSS
33%
10
NR Pre
NR Pre
NR Pre
NR Pre
R Pre
NR Pre
NR Pre
analysis has shown that empirical validations is higher, about
only 21% of the validation occurred in the academia
University of Maryland (UMLD) [3][19][22] and University School of
21% of the systems
that utilizes same data sets from public repository such as eclipse, NASA, and
projects studied are student’s
used in the validation of the relationship
O languages have

dominated the world of software applications. However, analysis indicate that about 54% of the
applications were written in C++ in both the industry and the academia while applications written
in Java is about 43% and 3% of the studies did
(See Fig. 6)
Fig
4.3.3. Study Type and Product Release
In the context of this study, study type refers to whether the study
Replicated studies were considered in this study because only few studies exist on empirical
validation of OO design metrics with respect to CK
replicated. Basili et al [6] has stressed the need for replicat
metrics, provide understanding and usefulness of the metrics with regar
faults. However, analysis shows that only 14% are replicated studies while 86% are non
replicated. This is captured in Fig. 7
Aggarwal et al [24] replicated Briand et al [22][27] and Zhou et al [33] replicated Olague et al
[23]. Other studies were also found reusing datasets of previous studies.
C++
53%
Non
Replicated
86%
in Java is about 43% and 3% of the studies did not mentioned the language of their application.
Figure 6.Programming languages used
4.3.3. Study Type and Product Release
In the context of this study, study type refers to whether the study is a replicated
validation of OO design metrics with respect to CK and SLOC while other
replicated. Basili et al [6] has stressed the need for replicated work as it assist to re
metrics, provide understanding and usefulness of the metrics with regard to different types of
However, analysis shows that only 14% are replicated studies while 86% are non
. This is captured in Fig. 7. Furthermore, Briand et al [22][27] replicated Basili et al[6],
[23]. Other studies were also found reusing datasets of previous studies.
Figure 7. Study type
Java
44%
Not_State
d
3%
Replicated
14%
Non-
Replicated
86%
0%
11
not mentioned the language of their application.
replicated one or not.
studies were
k as it assist to re-validate the
d to different types of
However, analysis shows that only 14% are replicated studies while 86% are non-
Furthermore, Briand et al [22][27] replicated Basili et al[6],

Based on the release type, we mean the state of the system studied when its structural properties
were measured and validated: pre
measuring of faults during development and testing, while those faults measured after the system
has been released to the users is the post
the systems used for the empirical validation where pre
the effectiveness of OO design metrics
addition, only 3% of the studies used post
faults at different levels of severity (High, M
4.4.Metrics Empirical Validation Usefulness
RQ4: Of what relevance are the empirical validations
considered, it has been shown that empirical evidences is a vital step
practical relevance of software metrics in software organizations. It
empirical evidence that product metrics are related to important external attributes like FP,
metrics will remain little or of no value. In
expression that depicts the theoretical basis for developing prediction models for relating OO
metrics and FP. The studies hypothesized that the relationship is due to the effects it has on
cognitive complexity. (See Fig.
impact on cognitive complexity which in turn, relates to FP. More
can lead OO classes exhibiting unwanted external qualities like FP
maintainability [10]. Thus, metrics that having the ability to measure these structural properties
are considered as good predictors of FP.
Figure 9.
Figure 8. Product release type
we mean the state of the system studied when its structural properties
were measured and validated: pre-release and post-release. For instance, pre-release means
ts during development and testing, while those faults measured after the system
has been released to the users is the post-release. However, analysis indicates that about 97% of
the systems used for the empirical validation where pre-release product. This findings confirmed
OO design metrics in evaluating the structural properties of OO classes. In
addition, only 3% of the studies used post-release application (maintenance) by categorizing
faults at different levels of severity (High, Medium and Low-impact errors) [12]. (See
4.4.Metrics Empirical Validation Usefulness
RQ4: Of what relevance are the empirical validations of software metrics?In all the studies
considered, it has been shown that empirical evidences is a vital step towards ensuring the
practical relevance of software metrics in software organizations. It indicates
metrics will remain little or of no value. In particular, studies by [2][10][16][21] provided an
cognitive complexity. (See Fig. 9) The indication is that the structural properties of classes have
impact on cognitive complexity which in turn, relates to FP. More so, high cognitive complexity
can lead OO classes exhibiting unwanted external qualities like FP, reduced understandability an
are considered as good predictors of FP.
ure 9. Theoretical basis of OO product metrics [10]
Pre-
release
97%
Post-
release
3%
12
we mean the state of the system studied when its structural properties
release means
ts during development and testing, while those faults measured after the system
release. However, analysis indicates that about 97% of
findings confirmed
evaluating the structural properties of OO classes. In
release application (maintenance) by categorizing
See Fig. 8)
In all the studies
towards ensuring the
that, without
[2][10][16][21] provided an
he indication is that the structural properties of classes have
, high cognitive complexity
reduced understandability and

13
The studies went further to explain that the expression of such a relationship can be used for early
prediction and identification of risky software classes or the construction of preventative (e.g.
design, programming) strategies [10]. To this end, using OO design metrics such as CK and
SLOC metrics can assist organizations to assess software development of any size swiftly at a
reduced cost, take solution actions early and thus, avoid costly rework [10][11][19][24].
4.5. Generic Metric for Fault Proneness Prediction
RQ5: Are there generic software metrics for predicting faulty classes?In this SLR, analysis have
shown that CK or CK and SLOC metrics have impact on class FP. Nonetheless, some studies did
not consider some of the metrics. Moreover, the results are contradicting even when same dataset
was used. For instance, in the study performed by [2] that utilized two systems written in C++
and Java, the results obtained indicates that WMC was significant with C++ but was not
significant with Java. Also, DIT was significant in few studies but insignificant in most studies.
This also applicable to other metrics. Fig. 10 presents the significance and insignificance
distribution of CK and SLOC metrics on FP of OO classes.
Figure 10. Validation of CK + SLOC relationship with FP
From the results obtained in this analysis, it is clear that there is no generic metric for FP, rather
best predictors of FP varies according to the type of applications used, the language used in
coding and the target application domain. In addition, SLOC, CBO, RFC, and WMC are the
metrics mostly reported as having strong significant relationship with FP in all the studies
followed by LCOM. This confirms the findings in [40][41]. In this case, the results were based on
the value of each metrics. Consequently, the higher the value, the higher the FP of the class.
Moreover, DIT and NOC were the metrics found to be mostly insignificant in all the studies
considered.
5. DISCUSSIONS
As OO programming has becomes the mainstream in software development today, several OO
metrics have been proposed and developed to assess the quality of OO software systems. By
assessing the quality of software during software development, quick design decisions at a
reduced cost can be ensured. With the 29 studies considered in this SLR, it shows that only few
empirical validation studies exist in the perspective of CK and SLOC metrics and FP prediction.
However, the studies considered used different OO measures such as coupling, cohesion,
inheritance and size measures to construct quality models that predicts the FP based on the
statistical techniques of LR, machine learning and so on. In addition, the predictive accuracy of
such models were reported based on either cross validation or goodness of fit [42]. Based on the
CBO RFC WMC LCOM DIT NOC SLOC
N/A 5 5 6 11 5 11 12
Insignificant 1 0 1 4 15 15 0
Significant 23 24 21 14 9 3 17
0
5
10
15
20
25
30
35
CK + SLOC Metric Validation

14
analysis conducted, LR is the most widely used model with high predictive accuracy as well as
the best in predicting faulty classes. These models utilizes FP as the dependent variable obtained
during the testing phase, while the OO metrics are the independent variables obtained during
design and coding phases. However, the statistical technique like LR can only predict the FP of a
class without giving information regarding the possible number of faults in that class.
Also, the study found that size, complexity, coupling measures were the metrics found to be
strongly related to FP followed by cohesion in the studies that considered CK+SLOC metrics.
Inheritance measures were found to be insignificant in several studies. This led some authors to
argued that DIT has an impact on the understandability of OO application and does not support
reusability, while others argued that the number of methods in the classes is the factor that affects
understandability [10][29]. With replicated studies, analysis shows that only few studies exist and
most of the studies were based on shared or reused dataset of previous studies obtained from
NASA, OSS (Mozilla, eclipse projects) and so on. Furthermore, results obtained from these
studies were not consistent in terms of significance level. Some metrics appears to be
significantly (positively or negatively) related to FP and some were not. Consequently, the best
predictors of FP depends on the type of language, applications and the targeted domain.This study
also found that the systems used in the empirical validation circled within the sphere of students,
OSS, and non-OSS projects which is the dominant of all. In addition, majority of the systems
were developed by professionals (79%). Also, validation were performed on only pre-release
products (97%) and only one (4%) study actually performed it on post-release product. However,
the study by [29] recommend that as a system evolves, it becomes more cumbersome to use OO
metrics to accurately identify the FP of classes in post-release products. To this end, alternative
methods needs to be applied if high accuracy is to be achieved. More so, only applications written
in C++ and Java were used to validate the relationship between OO metrics and FP.
The implication of this study is that empirical validation of OO metrics relationship with FP is
crucial to preserving the practical relevance of OO metrics in organizations. It can assist in the
quick allocation of resources to where they are needed most, avoid the costly rework and
facilitate other development activities such as change impact analysis, testing and so on.
Therefore, during development strong efforts have to be technically channelled to keeping all
those metrics at a reasonable level since FP of a class is based on each metric value.
5.1. Strengths and Weaknesses
This study covered at least large number of articles that assist in extracting relevant information
used. To this end, we are quite sure that the study actually covers the empirical validation of CK
and SLOC metrics published between January 1995 and December 2012. The SLR carefully
followed the guidelines by proffered by Kitchenham et al [26] where credible and trusted sources
were used. However, possible threats to this study could emanate from the search terms used, the
risks posed by not covering all the relevant studies or it could be that most relevant studies were
hidden in the excluded sources. Furthermore, threats could be the risk of misrepresenting the
findings of some of the papers found like not considering fault severity levels, positive or
negative significance of the metrics. Nonetheless, we have strong confidence that if such threats
exist, they have no significant effect on the results of this SLR. In this case, we worked
collaboratively, analysed all selected studies and all decisions as well as results were checked,
rechecked and inconsistencies resolved.
6. CONCLUSIONS
Today, as the OO paradigm has gained widespread popularity coupled with software
dependability, it is important that high software quality should not be compromised. OO design
metrics should always be used to assess software quality during software development. By this

15
evaluation, design quality will be improved which in turn would lower the probability of the
software being flawed. Doing this at the early phases of development can attracts a considerably
small cost and reduced efforts than late during development process. Several OO metrics have
been proposed in this direction like CK metric suite and the size measure. Moreover, many
empirical validation of the relationship between OO metric and FP have been reported. However,
to ascertain which of them are useful predictors of FP, this study explored the existing empirical
validation of CK+SLOC metrics to bring together the state-of-the-art in fault prediction using FP
and CK + SLOC metrics. The results obtained were presented and discussed.
The main findings of this SLR are as follows:
SLOC, CBO, RFC, WMC are metrics that strongly association with FP. There are also
considered the best predictors of FP in majority of the studies. Moreover, LCOM is
somehow an indicator of FP while DIT and NOC are found to be mostly insignificant.
With the results, we deduced that best predictors of FP depends on the class of
applications and the domain involved.
This study found 29 empirical studies that have validated CK and SLOC metrics with FP
of OO class. In these studies, 6 were from student’s project and 23 were from non-
students projects (mainly OSS and industrial applications).
Software applications written in C++ and Java were majorly used to empirically validate
the association between OO metrics and FP.
The prediction models constructed were mainly based on LR. Only few machine learning
and other techniques have been used. Thus, this study deduced that LR is the best
statistical technique used for FP prediction.
The empirical studies revolved around pre-release software products. Only one study has
performed empirical validation on post-release product.
Lastly, only few replicated studies exist. However, most studies were found reusing the
dataset of previous studies.
Future work will involve conducting systematic review on the empirical validation of the
relationship between FP and other OO metrics other than CK metric suite as well as
maintainability.
With the above findings, here are some recommendations:
a) To predict the FP with some level of accuracy using CK and SLOC metrics, SLOC,
CBO, RFC, WMC and LCOM are to be considered. Moreover, LR should be used as the
predictive model. Metrics such as DIT, and NOC should only be considered based on the
current value measured in that particular software product. This is because, though they
appears not to be regular FP indicators, however their significance or insignificance could
be as a result of either the developers’ experience or the inheritance strategy applied.
b) For high quality software to be ensured that is stable and maintainable, low-coupling,
highly cohesion, controlled size and inheritance should be adhered to.
c) For the evaluation of software quality during development or maintenance, measures
should strongly not be based on the nature of the environment involved, instead on steady
indicators of design problems and impacts on external quality attributes.
d) More empirical studies should be carried out on applications written in other OO
languages other than C++ or Java. Also, additional empirical studies should be performed
in the academia and more replicated studies should be carried out in order to re-validate
the metrics and keep them relevant.
e) More efforts should be channeled towards post-release software products in order to
confirm if models utilizing OO metrics can effectively predict class FP accurately or not.

16
f) During impact analysis of OO software systems, as a quality support activity, OO metrics
can be used to assess the software quality first before actual changes are made.
To this end, developers and maintainers should use these metrics consistently to evaluate and then
identify which OO classes requires attention in order to channel resources to those classes that are
likely failure in the field.
REFERENCES
[1] Xu, J., Ho, D. and Capretz, L.F (2008): An Empirical Validation of Object-Oriented Design Metrics
for Fault Prediction. Journal of Computer Science No.4, Vol 7, pp. 571-577. ISSN 1549-3636
[2] Subramanyam, R. and Krishnan, M.S.(2003): Empirical Analysis of CK Metrics for Object- Oriented
Design Complexity: Implications for Software Defects. IEEE Trans. Software Eng. No.29, pp. 297-
310,
[3] Zhou, Y., & Leung, H.(2006): Empirical analysis of object oriented design metrics for predicting high
and low severity faults. IEEE Transactions on Software Engineering, 32(10), pp. 771–784.
[4] Singh, Y. Kaur, A. and Malhotra, R.(2010): Empirical validation of object-oriented metrics for
predicting FP models. Software Quality Journal, vol.18 pp. 3–35.
[5] Succi, G., Pedrycz, W., Stefanovic, M., Miller, J.(2003): Practical assessment of the models for
identification of defect-prone classes in object-oriented commercial systems using design metrics.
Journal of Systems and Software 65, pp. 1-12.
[6] Basili, V., Briand, L., & Melo, W.(1996): A validation of object oriented design metrics as quality
indicators. IEEE Transactions on Software Engineering, No.22, Vol. 10, pp.751–761
[7] Chidamber, S., Darcy, D., Kemerer, C.(1998): Managerial use of metrics for object oriented software:
an exploratory analysis. IEEE Trans. Softw. Eng. Vol. 24, No. 8, pp. 629–639.
[8] Tang, M. H., Kao, M. H., & Chen, M. H. (1999): An empirical study on object-oriented metrics. In
Proceedings of 6th IEEE International Symposium on Software Metrics. pp.242–249.
[9] Janes, A. et al.(2006): Identification of defect-prone classes in telecommunication software systems
using design metrics. International Journal of Information Sciences.
[10] Emam, K.E., Melo, W.L., Machado, J.C.(2001): The prediction of faulty classes using object-oriented
design metrics. Journal of Systems and Software No. 56, pp. 63-75.
[11] Malhotra, R., Kaur, A. and Singh, Y.(2010): Empirical validation of object-oriented metrics for
predicting FP at different severity levels using support vector machines. International Journal System
Assurrance Engineering Management. No.1, vol. 3, pp. 269–281.
[12] Shatnawi, R. and Li, W.(2008): The effectiveness of software metrics in identifying error-prone
classes in post-release software evolution process. The Journal of Systems and Software no. 81,
pp.1868–1882.
[13] Boehm, B., Basili, V.(2004): Software Defect Reduction Top 10 Lists. IEEE Computer 34(1), 135–
137, 2001
[14] Myers, G., Badgett, T., Thomas, T., Sandler, C.(2004): The Art of Software Testing, second ed. John
Wiley& Sons, Inc., Hoboken, NJ.
[15] Kanmani et al (2007): Object-oriented software fault prediction using neural networks, Information
and software technology 49, 483-492.
[16] Chidamber, S., Kemerer, C.F.(1994): A metrics suite for object oriented design. IEEE Trans. Softw.
Eng. Vol. 20, No. 6, pp. 476–493.
[17] Yu, P., Systa, T., & Muller, H.(2002): Predicting FP using OO metrics: An industrial case study. In
Proceedings of Sixth European Conference on Software Maintenance and Reengineering, Budapest,
Hungary, pp.99–107.
[18] Fenton, N., Neil, M.(1999): Software metrics: successes, failures, and new directions. Journal of
Systems and Software vol. 47, pp. 149-157.
[19] Briand, L., Daly, J., Porter, V., & Wust, J.(1998): A comprehensive empirical validation of design
measures for Object Oriented Systems. Proceeding METRICS '98 Proceedings of the 5th International
Symposium on Software Metrics IEEE Computer Society, Washington, DC, USA.
[20] Fenton, N., Ohlsson, N.(2000): Quantitative analysis of faults and failures in a complex software
system. IEEE Transactions on Software Engineering, to appear.
[21] Emam, K.E., Benlarbi, S., Goel, N., Rai, S.N.(2001): The Confounding Effect of Class Size on the
Validity of Object-Oriented Metrics. IEEE Trans. Software Eng. 27, 630—650.

17
[22] Briand, L.C., Wüst, J., Daly, J.W., Porter, D.V.(2000): Exploring the relationships between design
measures and software quality in object-oriented systems. Journal of Systems and Software No. 51,
pp. 245—273.
[23] Olague, H.M., Etzkorn, L.H., Gholston, S., Quattlebaum, S.(2007): Empirical Validation of Three
Software Metrics Suites to Predict FP of Object-Oriented Classes Developed Using Highly Iterative
or Agile Software Development Processes. IEEE Trans. Software Eng. No.33, pp.402—419.
[24] Aggarwal, K. K., Singh, Y., Kaur, A. and Malhotra, R. Empirical Analysis for Investigating the Effect
of Object-Oriented Metrics on FP: A Replicated Case Study. Software Process Improvement and
Practice, No.14, pp. 39–62, 2009
[25] Gyimóthy, T., Ferenc, R., Siket, I.(2005): Empirical Validation of Object-Oriented Metrics on Open
Source Software for Fault Prediction. IEEE Trans. Software Eng. No.31, pp.897—910.
[26] Kitchenham, B. and Charters, S.(2007): Guidelines for performing Systematic Literature Reviews in
Software Engineering, Keele University and Durham University Joint Report, Tech. Rep. EBSE
2007-001.
[27] Briand, L.C., J. Wust and H. Lounis.(2001): Replicated case studies for investigating quality factors
in object-oriented designs. Empirical Software Engineering. No.6, pp. 11-58.
[28] Olague, H.M., Etzkorn, L.H., Messimer, S.L. and Delugach, H.S.(2008): An empirical validation of
object-oriented class complexity metrics and their ability to predict error-prone classes in highly
iterative, or agile, software: a case study. Journal of Software Maintenance. No. 20, pp.171-197.
[29] Rathore, S.S. and Gupta, A.(2012) Validating the Effectiveness of Object-Oriented Metrics over
Multiple Releases for Predicting FP. Proceedings of 19th Asia-Pacific Software Engineering
Conference, IEEE. pp.350-355.
[30] English, M., Exton, C., Rigon, I. and Cleary, B.(2009): Fault Detection and Prediction in an Open-
Source Software Project. In: 5th International Conference on Predictor Models in Software
Engineering.
[31] Goel, B. and Singh, Y.(2008): Empirical Investigation of Metrics for Fault Prediction on Object-
Oriented Software. Computer and Information Science, pp. 255-265.
[32] Shaik, A. et al.(2011): Investigate the Result of Object Oriented Design Software Metrics on FP in
Object Oriented Systems: A Case Study. Journal of Emerging Trends in Computing and Information
Sciences, Vol. 2 No.4, ISSN 2079-8407
[33] Zhou, Y., Xu, B. and Leung, H.(2010): On the ability of complexity metrics to predict fault-prone
classes in object-oriented systems. The Journal of Systems and Software No. 83, pp. 660–674.
[34] Al-Dallal, J. and Briand, L.C.(2010): An object-oriented high-level design-based class cohesion
metric. Information & Software Technology No. 52, pp.1346-1361.
[35] Al-Dallal, J.(2011): Transitive-based object-oriented lack-of-cohesion metric. Procedia Computer
Science, pp. 1581-1587.
[36] Pai, G.J., Dugan, J.B.(2007): Empirical Analysis of Software Fault Content and FP Using Bayesian
Methods. IEEE Trans. Software Eng. No. 33, pp.675-686.
[37] Johari, K. and Kaur, A.(2012):Validation of Object Oriented Metrics Using Open Source Software
System: An Empirical Study. ACM SIGSOFT Software Engineering Note, Vol. 37, No.1, pp.1.DOI:
10.1145/2088883.2088893
[38] Briand, L., Wuest, J., Ikonomovski, S. and Lounis, H.(1999): Investigating quality factors in object-
oriented designs: an industrial case study. In: Proceedings of the International Conference on
Software Engineering.
[39] Creswell, J.W.(2013): Research Design: Qualitative, Quantitative, and Mixed Methods Approaches,
4th Edition, ISBN-13: 978-1452226101
[40] Saxena, P. and Saini, M.(2011): Empirical Studies to Predict Fault Proneness: A Review.
International Journal of Computer Applications, Vol 22, ISBN: 0975 – 8887
[41] Khan, Y. A., Elish, M.O. and El-Attar, M.(2012): A Systematic Review on the Impact of CK Metrics
on the Functional Correctness of Object-Oriented Classes. Springer, pp. 258–273.
[42] Briand et al.(2002):Assessing the Applicability of Fault-Proneness Models Across Object-Oriented
Software Projects. IEEE Transactions of Software Engineering. Vol. 28, No. 7.
[43] Isong, B.E. and Ekabua, O.O. (2013) “A Systematic Review of the Empirical Validation of Object-
oriented Metrics towards Fault-proneness Prediction”. International Journal of Software Engineering
and Knowledge Engineering (IJSEKE) WSPC. Vol. 23, No. 10. pp. 1513–1540 DOI:
10.1142/S0218194013500484. ISSN: 0218-1940

18
AUTHORS
Dr. Isong, Bassey
Received B.Sc. degree in Computer Science from the University of Calabar, Nigeria
in 2004 and M.Sc. degrees in Computer Science and Software Engineering from
Blekinge Institute of Technology, Sweden in 2008 and 2010 respectively. Moreover,
he received a PhD in Computer Sciencein the North-West University, Mafikeng
Campus, South Africa in 2014. Between 2010 and 2014 he was a Lecturer in the
Dept. of Computer Science and Information Systems,University of Venda, South
Africa. Currently, he is a Lecturer in the Department of Computer Sciences,
Mafikeng Campus, North-West University. His research interests include Software Engineering,
Requirements Engineering, Software Measurement, Maintenance, Information Security, Software Testing,
Mobile Computing and Technology in Education.
Prof. Obeten, Ekabua
He is a Professor and Departmental Chair of the Department of Computer Science in
the Delta State University, Abraka, Nigeria. He holds BSc (Hons), MSc and PhD
degrees in Computer Science in 1995, 2003, and 2009 respectively. He started his
lecturing career in 1998 at the University of Calabar, Nigeria. He is the former chair of
the Department of Computer Science and Information Systems, University of Venda
and Department of Computer Science, North-West University, Mafikeng Campus,
South Africa. He has published several works in several journals and conferences. He
has also pioneered several new research directions and made a number of landmarks contributions in his
field and profession. He has received several awards to his credit. His research interest is in software
measurement and maintenance, Cloud and GRID computing, Cognitive Radio Networks, Security Issues
and Next Generation Networks.

STATE-OF-THE-ART IN EMPIRICAL VALIDATION OF SOFTWARE METRICS FOR FAULT PRONENESS PREDICTION: SYSTEMATIC REVIEW

More Related Content

What's hot (18)

Viewers also liked (17)

Similar to STATE-OF-THE-ART IN EMPIRICAL VALIDATION OF SOFTWARE METRICS FOR FAULT PRONENESS PREDICTION: SYSTEMATIC REVIEW (20)

Recently uploaded (20)

STATE-OF-THE-ART IN EMPIRICAL VALIDATION OF SOFTWARE METRICS FOR FAULT PRONENESS PREDICTION: SYSTEMATIC REVIEW