SOFTWARE CODE MAINTAINABILITY: A LITERATURE REVIEW

International Journal of Software Engineering & Applications (IJSEA), Vol.7, No.3, May 2016
DOI : 10.5121/ijsea.2016.7305 69
SOFTWARE CODE MAINTAINABILITY:
A LITERATURE REVIEW
Berna Seref and Ozgur Tanriover
Department of Computer Engineering, Ankara University, Ankara, Turkey
ABSTRACT
Software Maintainability is one of the most important quality attributes. To increase quality of a software,
to manage software more efficient and to decrease cost of the software, maintainability, maintainability
estimation and maintainability evaluation models have been proposed. However, the practical use of these
models in software engineering tools and practice remained little due to their limitations or threats to
validity. In this paper, results of our Literature Review about maintainability models, maintainability
metrics and maintainability estimation are presented. Aim of this paper is providing a baseline for further
searches and serving the needs of developers and customers.
KEYWORDS
Maintainability Model, Maintainability Metric, Maintainability Estimation, Maintainability Prediction
1. INTRODUCTION
Every software system needs to be modified in order to meet requirements of customers, users
and new technologies. Addition and deletion of codes and adopting the system to a new
operational platform are examples of the modification operations. ISO/IEC 9126 [1] defines
software maintainability as “the capability of the software product to be modified.” Another
definition [2] for software maintainability is as “the ease with which a software system or
component can be modified to correct faults, improve performance or other attributes, or adapt to
a changed environment.” On the other hand, maintainability also depends on the extend of use of
software constructs/patterns, programming paradigms/languages, application frameworks,
programming skill of developers, coding rules, design patterns etc.
It has been observed that maintenance effort in the software life cycle ranges from 65% to 75% of
total software development time [3]. As the majority of the time is spent in maintenance phase,
effort spent to increase maintainability effects software cost in a negative or positive way [4].
Hence, one of the most important aim of software engineers is developing maintainable software.
Maintainability feature of a software increases quality of it. With maintainable software, it is easy
to modify parts of the software, meet user or customer requirements in a shorter time and manage
the software efficiently.
Maintainability Index (MI) estimation calculation is one of the well-known maintainability
estimation techniques. However, MI estimation is still a problem. It is not clear that how MI

70
increases or decreases. There is sometimes no explanation of MI variation. For example, in Figure
1 [5], two versions of the same code are shown. They have different MI values.
Figure 1. Sample codes
In Figure. 1., while MI value of Sample 1 is 71, MI value of Sample 2 is 73. According to the MI
values, Sample 2 is more maintainable. On the other hand, this result is arguable. Count of line of
codes is smaller in Sample 1 and it is more readable but Sample 2 is more maintainable based on
MI values. There is a discussion about these samples at [5]. Contributions about this discussion
are listed below:
• An idea is that declaring all variables at the top (as in Sample 2) is reasonable but
minimizing an identifier's lifespan affects maintainability in a positive way. The counter
idea to it is that location of variables does not affect metrics.
• An idea counter to the sample 2. This idea supports that distance between the declaration of
the variable and where it is used is reduced, variable span is reduced. As a result,
maintainability is increased. However, given samples and results are opposite of this idea.
• An idea supports that MI difference between two samples are because of number of used
operator. Sample 2 has fewer operators that increase maintainability.
As it is seen, MI prediction and its accuracy are not clear. There are more different ideas that
support or not support usefulness of it on maintainability estimation. We think that Sample 1 is
more maintainable for the reason that it is more readable, understandable, effective and has a
fewer line of code.
Another discussion [6] is about MI values, their meanings and if they are dependent on
technology. In this discussion, MI values and their meanings are listed below:

71
• 85 and more: good maintainability
• 65-85: moderate maintainability
• 65 and below: difficult to maintain
In this discussion [6], common idea is it is not a good idea to use these numbers on different
technologies such as C# and Java. We agree with this common idea and we strongly believe that
these numbers can not show degree of maintainability on different technologies and can not
compare maintainability of them. For example, the same MI value of C and Java technologies can
not show that they have the same maintainability. C and Java codes have different syntax and
different terminology. While C code has pointers, Java code has generics. It is meaningless to
compare these two different technologies by using MI values.
A lot of researches have been carried out about maintainability, maintainability models,
maintainability metrics and maintainability prediction to increase maintainability of the software
and as a result, increase quality of the software and decrease software effort and cost. In these
researches; new models are proposed to predict maintainability of software at early stage of its
life cycle, factors which have positive or negative impact on maintainability are identified and
maintainability of software is measured. At these researches, new algorithms are proposed or
hybrid methods are tried to get more accurate results.
In this study, Literature Review about maintainability models, maintainability metrics and
maintainability estimation are presented to provide a baseline for further researches and to serve
the needs of developers and customers.
This paper is structured as follows: Section 2 illustrates review methodology. Results are given in
Section 3. Section 4 includes summary and discussions. Section 5 lists identified future works in
the literature. Section 6 illustrates limitation of this review. Finally, review is concluded in
Section 7.
2. METHOD
In this study, three stages which are defining research questions, designing the search strategy and
selecting the studies are followed.
2.1. Research Questions
In this literature review, we are trying to answer following research questions:
RQ1: Which models are proposed to evaluate maintainability?
Aim of RQ1 is identifying proposed models that evaluate and measure maintainability.
RQ2: Which methods or algorithms are carried out to propose models? At which rate they are
effective?
Aim of RQ2 is focusing on methods or algorithms that are used to propose a model and discuss
effectiveness of this model.

72
RQ3: Which metrics and methods used for maintainability estimation?
Aim of RQ3 is extracting metrics used for maintainability estimation and focusing methods to
estimate maintainability.
RQ4: Which metrics used to improve software maintainability?
Aim of RQ4 is researching usage of metrics to improve maintainability and effect of metrics on
maintainability.
2.2. Search Strategy
To limit the search to the most relevant search term, the listed steps are followed:
1. Major distinct terms are extracted from our research questions.
2. Our search term is updated with keywords.
3. Relevant papers are selected with the condition of having all keywords.
In step 1, candidate search terms are defined: maintainability, maintainability models,
maintainability evaluation, model effectiveness, maintainability estimation, maintainability
metrics, maintainability methods, improvement of maintainability. Then, keywords of the search
term are selected as “maintainability model”, “maintainability metric” and “maintainability
estimation.” Different spelling is accepted such as “maintainability metrics”, “model of system
maintainability”, “estimation of software maintainability” etc.
In step 3, papers which have all keywords in the same or different spelling are accepted for the
search.
Four electronic databases are included to search our review studies: IEEE – Xplore, Springer
Link, Science – Direct, and Wiley Online Library. Advanced search features of the each database
are used to get relevant literature.
Our search includes the period 2005 to 2015. Searching is carried out for conference papers and
journal papers separately for each database. Candidate papers are ordered from most relevant to
least relevant.
2.3. Study Selection
After getting candidate papers, it is aimed to select papers from the top five candidate papers if
they provide conditions to be selected. While selecting papers, for some searches, all of the top
five candidate papers are selected because of providing selection condition. On the other hand, in
some searches, we are not able to select any papers because of not providing selection condition.
Inclusion and exclusion criteria to select papers are listed below:
Inclusion criteria:
• The paper must take place at the first five candidate papers which are ordered from the
most relevant to the least relevant between the period of 2005 to 2015.

73
• The paper title must have search term. For example, in this study, search term is
constructed with two words. The paper’s title must include both of the words.
• Besides providing all conditions, abstract of the paper must be focused on “software” and
searching keywords.
Exclusion criteria:
• Studies does not take part at the first five candidate papers that are proposed by searching
engine.
• Studies does not have all keywords at their titles.
• Studies whose abstracts are not directly related to search area.
• Studies which are selected by us for this search before. (to prevent duplication)
After applying inclusion and exclusion criteria, relevant papers are narrowed to 20 and listed in
Table I with the features of ID, Authors, Addressed Research Questions and References No.
Table 1. Selected studies

74
Figure 2. Distribution of selected papers
0
2
4
6
8
10
12
14
IEEE Wiley
IEEE
Springer
ScienceDirect
Wiley
Figure 3. Distribution of selected studies based on electronic databases
We aimed to select up to date papers to be able to see the last level of maintainability researches.
Figure 2 gives information about distribution of selected studies per year. Figure 3 gives
information about number of selected studies based on electronic databases.
3. RESULTS
Firstly, it is observed that there are relatively few relevant studies matching our search terms
between 2005 and 2012. When Figure 2 is analyzed, it seen that number of studies about the
search term increases year by year and it becomes more active in 2013 and 2015. When Figure 3
is analyzed, it is observed that a large number of the selected studies is belong to IEEE while the
smallest part is belong to Wiley Online Library.
In addition, results show that 35% of the selected studies (7) are journal while the remaining 65%
of them are conference papers (13).
Review results for each research questions are given in the following subsections.

75
3.1. Proposed Maintainability Models
RQ1: Which models are proposed to evaluate maintainability?
I. Heitlager et al. (S1) discuss problems of Maintainability Index such as root cause analysis,
average complexity, computability comment, understandability and control. Then, they propose a
new model to measure maintainability. In this model, by using ISO 9126, source code measures
are mapped onto the sub characteristics of maintainability and a relationship between source code
metrics and ISO 9126 quality characteristics is provided.
F. Ye et al. (S2) propose a model based on multiple classifiers combination which has three parts:
attribute selection, model training and model interpretation.
O. Turetken (S3) discusses applicability of a software maintenance model and choses Software
Improvement Group Maintainability Model (SIG Model).
A. Kaur et al. (S4) discuss Oman and Hagemeister Maintainability Index model, show
ineffectiveness of it. After that, propose two regressions models built by stepwise selection and
backward elimination.
A. Kaur et al. (S5) propose “extensibility” as a sub character and evaluate aspect oriented
software maintainability quality using Analytic Hierarchy Process (AHP).
A.A. Moataz and A.H. Al-Jamimi (S6) propose fuzzy logic-based transparent prediction model.
A. Sheshasaayee and R. Jose (S7) propose aspect oriented maintainability model which is based
on static metrics for aspect oriented systems.
S. MISRA (S8) focuses on relationship between some software metrics and MI to increase
maintainability of a system.
K.T. Al-Sarayreh et al. (S9) propose a reference model of Software FUR (Functional User
Requirements) for identification of system maintainability requirements.
3.2. Methods/Algorithms and Effectiveness of Proposed Models
RQ2: Which methods or algorithms are carried out to propose models? At which rate they are
effective?
S1 decreases problems of MI with proposed model.
S2 uses Genetic Algorithm to select attributes (metrics) and then train the system with Weka tool.
After that, to interpret results of proposed model, uses rule extracting algorithm based on decision
tree. Dataset for this study is 300 classes of open source C++ software system downloaded from
internet. As a result, it is observed that the model trained by selected properties is faster than the
model trained by all properties. In addition, this model has a benefit of enabling developers and
researchers to select best maintainable object oriented software system.
S3 enables business process designers and practitioners to understand deficiencies and pinpoint
parts of the possible improvement places and predict efficiency of maintenance tasks. Doesn’t

76
taking into account cognitive physiology on the understandability of process models is one of the
limitation of this model.
S4 uses different versions of Luceno search engine which are 2.0.0 and 2.1.0 as a dataset and
calculates metrics of them by using CKJM and IntelliJ IDEA tool. Proposed regression models
are built by using Stepwise Selection and Backward Elimination. As a result, it is observed that
proposed models shows better performance when compared to Oman and Hagemeister (O&H)
Maintainability Index (MI) model based on correlation coefficient and Change between models.
Dataset used in this study is written in Java and it is seen as a limitation. As a future work, it is
planned to replicate the study on other software.
S5 evaluates aspect oriented software maintainability quality using Analytic Hierarchy Process
(AHP). The model consists of six attributes which are extensibility, reusability, modifiability,
analyzability, testability and modularity. Value of pair wise relative weights for the characteristics
are defined according to the survey that is carried out with 8 participants by taking mean of
survey results. Then, Eigenvector is computed to get relative ranking of the attributes in relation
to maintainability. As a result, it is observed that the order is: extensibility, reusability,
modifiability, analyzability, testability and modularity. In addition, it is observed that this model
can estimate maintainability of Aspect Oriented projects. In the future, it is planned to compare
maintainability of different Aspect Oriented projects with the weights which are gotten in this
study.
S6 predicts software maintainability using Mamdani fuzzy inference engine. QUES and UIMS
are selected as datasets. The model is compared with T-S-based, SVM (Support Vector Machine),
PNN (Probabilistic Neural Network), RBF (Radial Basis Function), BN (Bayesian Network) and
MARS (Multivariate Adaptive Regression Splines) models. As a result, it is observed that
Mamdani based model offers the best accuracy. Data scarcity is seen as a limitation for this study.
In the future, it is planned to try this model on different datasets.
S7 proposes Aspect Oriented maintainability model which is based on WOM (Weighted
Operations in Module), RFM (Response for a Module), CAE (Coupling on Advice Execution),
CDA (Crosscutting Degree of Aspect), CIM (Coupling on Intercepted Module), CFA (Coupling
on Field Access), CBM (Coupling between Modules), LCCO (Lack of Cohesion in Operation)
and LOCC (Lines of Class Code) metrics. In this study, aim is providing quality attributes to
predict parameters like maintainability, changeability etc. On the other hand, this model is
theoretical.
S8 focuses effect of 20 design/code measures on maintainability. These measures are extracted
using Krakatau metric professional measurement tool from the 50 projects written in C++
language. As a result, it is observed that there is a strong dependence between MI and SLOC
(Source Line of Code) metrics. There is a positive relationship between MI and AHF (Attribute
Hiding Factor), AIF (Attribute Inheritance Factor), AVPATHS (Average Depth of Paths), COF
(Coupling Factor), DIT (Depth of Inheritance Tree), MIF (Method Inheritance Factor) and RFC
(Response for Classes) metrics. There is a strong effect of ACLOC (Average Class Size),
AMLOC (Average Method Size), AVPATHS (Average Depth of Paths), CDENS (Control
Density), COF (Coupling Factor), DIT (Depth of Inheritance Tree), n (Program Vocabulary), N
(Program Length), PPPC (Percentage Public/Protected Members), and WMC (Weighted Methods

77
in Classes) metrics on MI. However, the results are gotten using 50 systems so results are not
universally general for all systems. In the future, it is planned to carry out further theoretical and
experimental studies with the aim of validating the obtained results in this study.
S9 aims to provide a model to identify system maintainability requirements and proposes
Software FUR (Functional User Requirements) model to achieve this aim. As a result, with this
study, budget overruns and missed deadlines are prevented, input requirements are determined
earlier in the project life cycle. However, in this study, only software requirements are allocated,
and the study is not expanded to hardware and manual system requirements.
3.3. Metrics and Methods for Maintainability Estimation
RQ3: Which metrics and methods used for maintainability estimation?
H. Washizaki (S10) proposes CCOF (Component Coupling Factor) metric which takes into
consideration of characteristics of remote components to measure coupling-based complexity of
Component Based Software (CBS) system. As a result, it is observed that CCOF is a useful and
effective index for maintainability. In the future, it is planned to expand number of samples.
L. Kumar et al. (S11) predict maintainability of Quality Evaluation System and User Interface
System using Neuro Genetic Algorithm. Genetic Algorithm is used to find optimum weights in
the Neural Network. Metrics are inputs for the network that will be trained. Mean Absolute Error
(MAE), Mean Absolute Relative Error (MARE), Root Mean Square Error (RMSE) and Standard
Error of the Mean (SEM) are performance evaluation criterions. As a result, it is observed that
this method shows better performance when compared to the last studies. On the other hand, this
model is created using object oriented systems so the model is likely to be valid for the systems
which are developed using object oriented programming languages. In the future, it is planned to
entegrate other techniques such as Particle Swarm Optimization, Fuzzy Logic, Clonal Selection
Algorithm to the network to increase accuracy rate of estimation.
Z.Yuming and X. Baowen (S12) study relationship between a number of metrics and
maintainability, and prediction ability of these metrics for software maintainability. 148 Java
open source software collected from the websites http://sourceforge.net/ and http://java-source.net
and used as dataset. 15 design metrics are extracted from the dataset and prediction ability of
these metrics are reported when used together. As a result, it is seen that average control flow
complexity per method (OSAVG) is the most important maintainability factor while cohesion and
coupling are less. In addition, multivariate prediction model is showed a good accuracy.
A. Pratap et al. (S13) predict software maintainability using Fuzzy Logic on Matlab Platform. In
this study, while Adaptability (AD), Complexity (CLX), Understandability (USD),
Documentation Quality (DocQ) and Readability (RD) metrics are used as input, maintainability is
used as an output. Triangular Membership Function, Trapezoidal Membership Function and
Gaussian Membership function are chosen membership functions. As a result, it is observed that
this model can be used to predict software maintainability. It is advised that to increase
maintainability of the system, Adaptability (AD), Understandability (USD), Documentation
Quality (DocQ) and Readability (RD) metrics should be high while Complexity (CLX) should be
low. As a future work, it is planned to propose new metrics and new approaches for software
maintainability.

78
H.S. Chae et al. (S14) study relationship between LOC (Line of Code), LCOM (Lack of Cohesion
in Methods), RFC (Response for Classes), DAC (Data Abstraction Coupling) metrics and
software maintenance effort. Dataset is constructed with 4 developers by recording the
maintenance time needed for each class of maintenance action. Correlation analysis is carried out
to observe relationship between the selected metrics and maintenance effort. As a result, it is seen
that selected metrics can not show a good performance for Web-based applications to predict
maintenance effort. In the future, it is planned to propose new metrics that take into account
characteristics of design patterns of Web based applications.
3.4. Metrics to Improve Software Maintainability
RQ4: Which metrics used to improve software maintainability?
N. Yoshida et al. (S15) propose dividing source code into functional segments using cohesion
metric to improve software maintainability. As a result, start and end points of each functional
segment are defined; understandability of the source code is increased. In this study, NCOCP2
(Normalized with number of LCOM (Lack of Cohesion in Methods)) metric is calculated and a
threshold value is set to NCOCP2. Then, functional segments are extracted based on this cohesion
metric. In the future, it is planned to validate the cohesion metric theoretically and propose a
method which add feature names to extracted functionalities.
F. Zhang et al. (S16) focus the effects of six context factors on distribution of software
maintainability metrics. These context factors are application domain, programing language, age,
life span, number of changes and number of downloads. As a dataset, 320 nontrivial software
systems from Source Forge is used. Kruskal Wallis test, Mann-Whitney U test and Cliff’s δ effect
size are selected as statistical methods to analyze 320 software systems. As a result, it is observed
that all context factors affect 39 calculated maintainability metrics. But the most are: application
domain, programming language and number of changes. In the future, it is planned to use more
software system from SourceForge, GoogleCode, and GitHub.
S. Counsell et al. (S17) study relationship between MI and coupling, defects and size features.
Three releases of two Eclipse projects are used as datasets. Class-based metrics are extracted with
JHawk tool. As a result, it is observed that there is a relationship between the features and MI.
D. Baski and S. Misra (S18) propose data weight of a web service description language, distinct
message ratio metric, message entropy metric and message repetition scale metric to evaluate
maintainability of XML Web Service. All metrics are evaluated using Weyuker’s properties and
Caner’s framework. As a result, usefulness of the metrics are proved. In the future, it is planned to
develop an automated tool to compute metrics and make researches to assign right values for
upper and lower boundaries of the complexity values for proposed metrics.
J.M. Conejero et al. (S19) try to find an answer for the question that if a certain crosscutting
characteristics affect software maintainability. Correlation between crosscutting properties and
changeability and stability attributes is focused. As a result, it is observed that certain crosscutting
properties affect changeability and stability in a negative way. In the future, it is planned to carry
out some empirical studies to compare obtained results.
J. de A.G. Saraiva et al. (S20) propose metrics’ categorization to define which metrics can be
used in the experiments to increase rate of maintainability. 7 categories and 17 subcategories are
presented. These are tested using Wilcoxon Test and survey which includes 47 expert opinion

79
about the proposed catalog. As a result, it is observed that proposed approach is useful and has a
positive effect on maintainability. In the future, it is planned to improve catalog generalization
and entegrate GQM (Goal-Question Metric) Model to the study.
4. SUMMARY AND DISCUSSIONS
To achieve maintainability evaluation and estimation, in general, databases and software metrics
are needed. Databases are obtained in two ways: constructing a new databases for the study or
using ready databases. Databases and tools to extract metrics used in the literature are listed in
Table II [8,10,14,17,18,22,23].
Table 2. Databases and tools
Database Tools
300 classes of open source C++ software
system downloaded from internet
Weka
Luceno search engine CKJM and IntelliJ IDEA tool
User Interface Management System
(UIMS) contains 39 classes and Quality
Evaluation System (QUES) contains 71
classes
-
50 projects written in C++ language
downloaded from several websites
static analysis tool
148 Java open source software collected
from the websites http://sourceforge.net/
and http://java-source.net
-
320 nontrivial software systems from
Source Forge
https://bitbucket.org/serap/contextstudy
http://www.scitools.com
Three releases of two Eclipse projects JHawk tool
In Table 2, it is observed that projects that is used as dataset are written in Java or C++. It can be
said that researchers preferred using Java or C++ projects instead of C# or C projects.
As it is seen in the literature, some studies are focused on problems of MI and tried to decrease
these problems with proposing new models and methods and analyzing relationship between MI
and software metrics.
MI is a measure that define how maintainable a system is. Higher values of MI means more
maintainable system. MI was introduced in 1992 by Paul Oman and Jack Hagemeister and
presented at the International Conference on Software Maintenance ICSM 1992 and formulated
in Equation (1) [27].
171-5.2In(HV)-0.23CC-16.2In(LOC)+50.0sin√2.46*COM
(1)
In Equation (1), HV is Halstead’s Volume, CC is McCabe’s cyclomatic complexity, LOC is line
of code, COM is percentage of comments.
Problems of MI can be listed as follows:

80
• It is not known that why sin operation is used, why CC is multiplied with 0.23 value or how
we can get 171. There are some understandability questions about MI formula but there is
no clear explanation.
• In MI Formula, COM takes place. However, it is one of the arguable points if comments
must take part in the formula.
• MI Formula can mask the presence of high-risk parts especially for object oriented systems
because of power low distribution of complexity and as a result, giving low average
complexity. It is not advised to use MI Formula as a maintainability measure for object
oriented systems.
• HV, CC, LOC and COM are used in MI Formula according to the statistical correlation.
However, there is no clear explanation and strong evidence that there is a causal relation
between the metrics. Because of it, it is hard to tune maintainability index by changing used
metrics in the Formula.
•
To increase performance of MI, some metrics are proposed and relationship between MI and
metrics are analyzed. Relationship between MI and metrics are shown in Table 3 [14,23].
Table 3. Relationship between MI and software metrics
Metric Relationship
AMLOC (Average method size) non-linear correlation
SLOC (Source lines of code) non-linear correlation
AHF (Attribute hiding) positive relationship
AIF (Attribute inheritance
Factor)
positive relationship
AVPATHS (Average depth of paths) positive relationship
COF (Coupling) positive relationship
DIT (Depth of inheritance tree) positive relationship
MIF (Method inheritance) positive relationship
RFC (Response for class) positive relationship
FIN (Number of incoming couplings) Correlation is not significant at the 5%
level or below according to Spearman’s
and Pearson’s coefficient
CBO (Coupling between objects) Correlation is significant at the
1% or 5% level according to Spearman’s
NOS (Number of java statements in a
class)
Correlation is significant at the
1% or 5% level according to Spearman’s
As it is seen in Table 3, there is a non-linear correlation between MI and AMLOC and SLOC.
When AMLOC is increased, method size increases, method becomes more complex and rate of
readability decreases, as a result, maintainability decreases. When SLOC is increased, program

81
size and complexity increases as a result, maintainability decreases. According to the coefficient
of correlation analysis, when attribute hiding, attribute inheritance, method inheritance, average
depth of paths, coupling, depth of inheritance tree and response for class are increased,
maintainability increases. In addition, according to the experiment which is carried on with two
releases of two projects, correlation rate between MI and CBO and NOS are significant at the 1%
or 5% level according to Spearman’s and Pearson’s coefficient but not significant at the 5% level
or below for FIN.
There are some metrics used to predict maintainability of the systems. These metrics, systems and
performance of the metrics are listed in Table 4 [12,16,18-20].
Table 4. Metrics for maintainability estimation
Metric System Performance
CCOF(Component coupling
factor)
Component Based
System (CBS)
useful index for
maintainability, nonredundant
with existing metrics
DIT (Depth of the
inheritance tree)
UIMS dataset
QUES dataset Both
dataset (Merging
UIMS and QUES
Dataset)
Strong Pearson’s correlation
coefficient (-0.43) with
dependent variable CHANGE
(counting the number of lines
in the code which has been
changed during a 3-year
maintenance period) for UIMS
Dataset.
NOC (Number of children) UIMS dataset QUES
dataset Both dataset
(Merging UIMS and
QUES
Dataset)
coefficient (0.56) with
Dataset.
MPC (Message-passing
coupling)
QUES dataset
UIMS dataset Both
dataset (Merging
UIMS and QUES
Dataset)
maintenance period) for QUES
Dataset.
RFC (Response for a class) UIMS dataset QUES
(Merging UIMS and
QUES
Dataset)
Dataset.

82
LCOM (Lack of cohesion of
methods)
UIMS dataset QUES
(Merging UIMS and
QUES
Dataset)
Dataset.
DAC (Data abstraction
coupling)
UIMS dataset QUES
(Merging UIMS and
QUES
Dataset)
Dataset.
WMC (Weighted method
per class)
Both dataset
(Merging UIMS and
QUES
Dataset)
UIMS dataset
QUES dataset
maintenance period) for both
datasets.
NOM (Number of
methods)
UIMS dataset QUES
(Merging UIMS and
QUES
Dataset)
Dataset.
SIZE1 (Lines of code) Both dataset
(Merging UIMS and
QUES
Dataset)
UIMS dataset QUES
dataset
maintenance period) for both
dataset.
SIZE2 (Number of
properties)
UIMS dataset QUES
(Merging UIMS and
QUES
Dataset)
dataset
OSAVG (Average 148 Java open OSAVG is the most predictive

83
complexity per method)
CSO (Average number of
methods per class)
CSA (Average number of
attributes per class)
SNOC (Average number of
children per class)
source software metric for maintainability
based on multivariate linear
regression model. CSO and
CSA follows it.
AD (Adaptability) CLX
(Complexity)
USD (Understandability)
DocQ (Documentation
Quality)
RD (Readability)
Some values are
assumed for the
metrics
When these metrics are used as
input and maintainability is
used as output, maintainability
can be predicted based on
fuzzy logic.
AD, USD, DocQ and RD
should be high whereas CLX
should be low to improve
maintainability.
RFC (Response for classes)
LCOM (Lack of cohesion in
Methods)
DAC (Data abstraction
coupling)
LOC (Line of code)
Web-based
applications
These metrics are not able to
relate with maintenance for
Web-based applications.
In Table 4, metrics used for maintainability estimation, where they are used and their prediction
performance are listed. It is observed that, some metrics affect maintainability in a positive way
whereas some of them affect negative. In addition, according to this table, it is important to
choose right metrics for the system. For example, in the table, there are some metrics used for
maintainability estimation of Web-based applications but these metrics fail to predict.
Some approaches are tried to improve maintainability of the systems such as using metrics,
analyzing effect of context factors on metrics etc. These approaches are listed with aim of them
and results in Table 5 [21-22].
Table 5. Approaches to improve maintainability
Approach Aim Result
NCOCP2 (Normalized
with Number of Lock of
Cohesion in Methods)
cohesion metric is used.
This metric is used to
divide source code into
functional segments with
the aim of identifying
start and end points of
each functional segments.
Functional segments are
identified. Understandability
of source code is increased,
as a result, maintainability is
increased.
Effect of context factors
(application domain,
programming language,
age, life span, number of
changes, number of
Effect of six context
factors on software
maintainability metrics is
analysed to improve
maintainability.
It is observed that,
application domain,
programming language and
number of changes effect
software maintainability

84
downloads) on
distribution of software
maintainability metrics
are analysed
metrics mostly.
To predict or evaluate maintainability of a system a lot of methods and correlation analysis are
used. Fuzzy Logic, Mamdani Fuzzy Inference Engine, Genetic Algorithm, regression models
built by stepwise selection and backward elimination, classical linear regressions, Kruskal Wallis
test, Mann-Whitney U test, Cliff’s δ effect size and statistical methods are one of the most used
methods to predict or evaluate maintainability of a system.
Maintainability is an important point for software architecture and software design decisions. A
lot of studies are carried out about maintainability of software architecture however; they are not
able to analyze the optimal maintainability of a software architecture very well [28]. Design takes
part in the software architecture development process. Architectural design decisions are taken at
the beginning of project and these decisions affect software maintainability in a significant rate.
For example; merging two components can affect maintainability in a negative way [29]. In the
future, it is needed to focus on effect of design decisions on maintainability and assess
maintainability to the software architecture to improve performance of the systems.
5. FUTURE WORKS IDENTIFIED IN REVIEWED PAPERS
Future works identified in reviewed papers are listed below:
• S1 aims to investigate if their rating schemas could be captured using Bayesian Belief
Nets (BBN) and incorporate ISO 25000 series (SQuaRE) with their proposed
maintainability model.
• S2 aims to take consideration of other object-oriented metrics and increase number of
projects.
• S3 aims to take consideration of the role of cognitive physiology on the understandability
of the models.
• S4 aims to apply the proposed model on other software and validate the model.
• S5 aims to compare maintainability of various Aspect Oriented projects using obtained
weights in the study.
• S6 aims to take consideration of uncertainty which means while some internal attributes
affect some projects in a positive or negative way, these attributes can not affect all of
the projects.
• S8 aims to study the effects of several factors that analysis methods used in this study
depend on. In addition, it is planned to validate the results obtained in this study by using
further theoretical and experimental studies.
• S9 aims to evaluate proposed approach using a number of case studies.
• S10 aims to verify the experimental result and modify the component specification step
of CCOF metric.
• S11 aims to entegrate other techniques such as Particle Swarm Optimization, Fuzzy
Logic and Clonal Selection Algorithm to the neural network to increase accuracy rate of
estimation.
• S13 aims to add new metrics to the system for software maintainability.

85
• S14 aims to apply the study to other web based applications and propose new metrics for
Web-based applications.
• S15 aims to validate the cohesion metric theoretically and propose a method which add
feature names to extracted functionalities.
• S16 aims to use more software system from SourceForge, GoogleCode and GitHub and
derive the thresholds and ranges of metric values according to the results obtained in the
study.
• S18 aims to develop an automated tool to compute metrics and make researches to assign
right values for upper and lower boundaries of the complexity values for proposed
metrics.
• S19 aims to carry out some empirical studies to compare obtained results.
• S20 aims to improve catalog generalization and entegrate GQM (Goal-Question Metric)
Model to the study.
6. LIMITATIONS OF THIS REVIEW
In this study, 4 keywords are searched using 4 search engines. However, number of keywords and
search engines can be increased. Also, different synonyms can be tried to increase number of
papers. For example, in this study, “maintainability estimation” keyword is used to get papers
which are about maintainability estimation. On the other hand, with this keyword, we are not able
to get the papers with title “maintainability prediction” even if they have the same content. Deep
searches can be carried out in the future.
7. CONCLUSION AND FUTURE WORK
In this study, results of a Literature Review about maintainability models, maintainability metrics
and maintainability estimation are presented to provide a baseline for further researches and to
serve the needs of developers and customers.
As a future work, it is planned to increase number of search engine and number of keywords to
get more relevant journals and conferences, focus on effect of design decisions on maintainability
and assess maintainability to the software architecture to improve maintainability.
REFERENCES
[1] ISO/IEC, ISO/IEC 9126. Software Engineering – Product quality 6.5.ISO/IEC, 2001.
[2] IEEE Std. 610.12-1990. 1993. Standard Glossary of Software Engineering Terminology, IEEE
Computer Society Press, Los Alamitos, CA, 1993.
[3] S. Muthanna, K. Konotogiannis, K. Ponnambalam, and B. Stacey, “A maintainability model for
industrial software systems using design level metrics,” In Seventh Working Conference on Reverse
Engineering, pages 248–256, November 2000.
[4] B. Kumar, “A Survey of Key Factors Affecting Software Maintainability,” International Conference
on Computing Sciences, DOI 10.1109/ICCS.2012.5.
[5] http://stackoverflow.com/questions/2749082/why-does-this-maintainability-index-increase
(16.01.2016)
[6] http://stackoverflow.com/questions/592866/maintainability-index (16.01.2016)
[7] I. Heitlager, T. Kuipers and J. Visser,”A Practical Model for Measuring Maintainability”, Sixth
International Conference on the Quality of Information and Communications Technology, pp. 30-39.

86
[8] F.Ye, X. Zhu, Y. Wang, “A New Software Maintainability Evaluation Model Based on Multiple
Classifiers Combination,” 2013 International Conference on Quality, Reliability, Risk, Maintenance,
and Safety Engineering (QR2MSE), pp. 1588-1591.
[9] O. Turetken, “Towards a maintainability model for business processes: Adapting a software
maintainability model (position paper),” 2013 IEEE 1st International Workshop on Communicating
Business Process and Software Models Quality, Understandability, and Maintainability (CPSM), pp.
1 - 4.
[10] A. Kaur, K. Kaur, K. Pathak, “A proposed new model for maintainability index of open source
software,” 2014 3rd International Conference on Reliability, Infocom Technologies and Optimization
(ICRITO) (Trends and Future Directions), pp. 1 - 6.
[11] A. Kaur, P.S. Grover, A. Dixit, “Quantitative evaluation of proposed maintainability model using
AHP method,” 2015 2nd International Conference on Computing for Sustainable Global
Development (INDIACom)”, pp. 1367 – 1371, 2015.
[12] M.A. Ahmed, H.A. Al-Jamimi, “Machine learning approaches for predicting software maintainability:
a fuzzy-based transparent model,” Software, IET, Volume: 7, issue: 6, pp. 317 - 326, DOI:
10.1049/iet-sen.2013.0046.
[13] A. Sheshasaayee, R. Jose, “A Theoretical Framework for the Maintainability Model of Aspect
Oriented Systems”, The 2015 International Conference on Soft Computing and Software Engineering
(SCSE 2015), Procedia Computer Science 62 (2015) 505 – 512.
[14] S. MISRA, “Modeling Design/Coding Factors That Drive Maintainability of Software Systems,”
Software Quality Journal, 13, 297–320, 2005.
[15] K. T. Al-Sarayreh, A. Abran and J. J. Cuadrado-Gallego, “A standards-based model of system
maintainability requirements,” J. Softw.: Evol. and Proc. 2013; 25:459–505.
[16] H.Washizaki, T. Nakagawa, Y. Saito, Y. Fukazawa, “A Coupling-based Complexity Metric for
Remote Component-based Software Systems Toward Maintainability Estimation,” 13th Asia Pacific
Software Engineering Conference (ASPEC 2006), pp. 79 - 86.
[17] L. Kumar, D. Kumar Naik, S. Ku. Rath, “Validating the Effectiveness of Object-Oriented Metrics for
Predicting Maintainability,” Third International Conference on Recent Trends in Computing (ICRTC’
2015), Procedia Computer Science 57 (2015) 798 – 806.
[18] Y. Zhou and B. Xu, “Predicting the Maintainability of Open Source Software Using Design Metrics,”
Wuhan University Journal of Natural Sciences, Vol.13 No.1, 014-020.
[19] A. Pratap, R. Chaudhary, K. Yadav, “Estimation of software maintainability using fuzzy logic
technique,” 2014 International Conference on Issues and Challenges in Intelligent Computing
Techniques (ICICT), pp. 486 - 492.
[20] H.S. Chae, T.Y. Kim, W.S. Jung, J.S. Lee, “Using Metrics for Estimating Maintainability of Web
Applications: An Empirical Study,” pp. 1053 - 1059.
[21] N. Yoshida, M. Kinoshita, H. Iida, “A cohesion metric approach to dividing source code into
functional segments to improve maintainability,” 2012 16th European Conference on Software
Maintenance and Reengineering (CSMR), Pages: 365 - 370.
[22] F. Zhang, A. Mockus, Y. Zou; F. Khomh, A.E. Hassan, “How Does Context Affect the Distribution
of Software Maintainability Metrics?,” 2013 29th IEEE International Conference on Software
Maintenance (ICSM), pp. 350 - 359.
[23] S. Counsell X. Liu, S. Eldh, R. Tonelli, M. Marchesi, G. Concas, A. Murgia, “Re-visiting the
'Maintainability Index' Metric from an Object-Oriented Perspective,” 2015 41st Euromicro
Conference on Software Engineering and Advanced Applications (SEAA), pp. 84 - 87.
[24] D. Baski and S. Misra, “Metrics suite for maintainability of eXtensible Markup Language web
services,” Software, IET, Volume: 5, issue: 3, pp. 320 - 341.
[25] J. M. Conejero, E. Figueiredo, A. Garcia, J. Hernandez, E. Jurado, “On the relationship of concern
metrics and requirements maintainability,” Information and Software Technology 54 (2012) 212–238.
[26] J.A.G. Saraiva, M.S. de França, S. C.B. Soares, F. J.C.L. Filho, R. M.C.R. de Souza, “Classifying
metrics for assessing Object-Oriented Software Maintainability: A family of metrics’catalogs,” The
Journal of Systems and Software 103 (2015) 85–101.
[27] http://avandeursen.com/2014/08/29/think-twice-before-using-the-maintainability-index/ (26.12.2015)
[28] http://www.janbosch.com/Articles/OptimalSAMaintainabilityCSMR01.pdf (27.15.2015)

87
[29] F. Oquendo, B. Warboys, R. Morrison, Software Architecture: First European Workshop, EWSA
2004, St Andrews, UK, May 2004, Proceedings, B. Graaf, “Maintainability through Architecture
Development” pp:206 -2010.
Authors
Berna Seref received the B.S. degree in computer engineering from Anadolu University, Eskisehir,
Turkey, in 2012 and the M.S. degree in computer engineering from Kirikkale University, Kirikkale,
Turkey, in 2015. She is currently pursuing the Ph.D. degree in computer engineering at Ankara University,
Ankara, Turkey and working as a research assistant.
Ozgur Tanriover received the M.S. and PhD. degrees in information systems from Middle East Technical
University, Ankara, Turkey in 2002 and in 2008 respectively. Since 2012, he has been working as an
assistant professor in computer engineering department at Ankara University.

SOFTWARE CODE MAINTAINABILITY: A LITERATURE REVIEW

More Related Content

What's hot (20)

Similar to SOFTWARE CODE MAINTAINABILITY: A LITERATURE REVIEW (20)

Recently uploaded (20)

SOFTWARE CODE MAINTAINABILITY: A LITERATURE REVIEW