Empirical Evaluation of Reusability Models

Andreea Cristina Lung, Simona Motogna

and Vladiela Petras¸cu

Babes-Bolyai University, Department of Computer Science, M. Kogalniceanu Street, Cluj-Napoca, Romania

Keywords:

Software Quality, Reusability, Empirical Study.

Abstract:

Many research efforts have been directed into ways to quantify the degree to which a software component can

be reused in other systems. As such, a number of different reusability assessment models have been proposed

in the literature, taking into account several metrics that can affect reusability and different approaches to

measuring it. In this paper, we conduct a longitudinal reusability assessment by applying three reusability

models to a number of different projects (libraries and frameworks) with the goal of studying the long-term

evolution of reusability in open-source software. The exploratory part of the study consists of reproducing and

applying the chosen models on three different-sized projects for several released versions of the software and

studying the transformations of reusability over time and how these relate to certain changes in quality factors

or size of the software. Results show a more intense variation of reusability in earlier versions and more stable

values towards later versions in applications, and a clear inﬂuence of complexity, modularity and cohesion on

reusability scores.

1 INTRODUCTION

In software development, reusability established itself

as a vital aspect seeked by many organizations or in-

dividual developers in order to speed up their devel-

opment times, increase productivity and boost perfor-

mance of their teams. Reusability also has a positive

impact on maintainability and overall quality of the

ﬁnal products. Regardless if the used code is an ex-

ternal library or a dependency for the project or a sim-

ply inserted third party component into a project, it is

clear that software reuse is heavily used in one form

or another.

Open-source code sharing platforms include li-

braries, frameworks and other assets easy to be

reused. Finding the right component to integrate in

a software system, on one hand, and preserving the

reuse potential of developed code, on the other hand,

might be challenging tasks. As a result, measuring the

reusability level and monitoring it becomes an impor-

tant aspect of such software systems.

In order to correctly quantify reusability, it is im-

portant to differentiate it from reuse. Literature de-

ﬁnes reuse as the use of existing artifacts or knowl-

edge to create new software (Frakes and Terry, 1996),

and reusability as the extent to which a program can

https://orcid.org/0000-0002-8208-6949

https://orcid.org/0000-0002-7878-4349

be reused in other applications (McCall et al., 1977)

or, more generally, the degree to which an asset can

be used in more than one system, or in building other

assets (ISO/IEC 25010, 2011). While reuse has es-

tablished itself as an activity that brings many sav-

ings to software development, reusability metrics aim

to identify the assets that will facilitate those savings

(Poulin, 1994).

Especially for software systems that are specif-

ically created for reuse purposes, keeping a track

record of how reusability evolves across the develop-

ment of the project can become a very tedious, but

signiﬁcant task, as any moment of low consideration

towards software quality could lead to a decrease in

the reusability level, hence decreasing the re-use po-

tential of the library or framework altogether and af-

fecting how people who use it interact with it or driv-

ing away new potential users of that library. For this

reason, factors affecting reusability should also be

tracked during development.

Therefore, keeping the reusability level under con-

trol should be a goal that any such system should

strive to achieve, and we aim to facilitate this process

and provide a starting guideline through the current

empirical study. We aim to investigate how reusabil-

ity evolves in large software systems over signiﬁcant

periods of time, how it is inﬂuenced by characteristics

of the code and which is the relation with the main-

Lung, A., Motogna, S. and Petra¸scu, V.

Empirical Evaluation of Reusability Models.

DOI: 10.5220/0011143100003266

In Proceedings of the 17th International Conference on Software Technologies (ICSOFT 2022), pages 265-275

ISBN: 978-989-758-588-3; ISSN: 2184-2833

265

tainability of the system.

The core contribution of our work lays in the fact

that we study reusability in the context of software

evolution, namely by investigating several versions of

the software systems. The second contribution lies

in the analysis of several reusability estimation mod-

els across the whole release cycle of the case study

projects.

The rest of the paper is organized as follows. We

start with presenting related work, in order to state

the contribution brought by our investigation. In Sec-

tion 3, we explain how we design our exploratory case

study, formulating the research questions and describ-

ing the setup of the case study. Section 4 presents our

ﬁndings in form of responses to the research questions

and draw some lessons learned. Then, we identify the

main threats to validity and discuss how we mitigate

them. We end by drawing some conclusions and de-

signing future research directions.

2 RELATED WORK

Reusability has been a research topic since its intro-

duction in the ﬁrst software quality model, proposed

by McCall in 1977 (McCall et al., 1977). Several

models for estimating the degree to which a software

asset can be reused have been proposed and a num-

ber of comparative studies address the differences be-

tween them.

In (Poulin, 1994), the author summarizes various

existing reusability estimation approaches and asso-

ciated metrics, based on a taxonomy distinguishing

between empirical and qualitative ones. While empir-

ical approaches use objective, quantiﬁable software

attributes as a basis for reusability metrics, qualita-

tive ones rely on offering guidelines and assigning

reusability estimators to components, based on com-

pliance to those guidelines. The paper emphasizes

the fact that, despite the considerable amount of ap-

proaches proposed in the literature - 11 empirical and

3 qualitative ones being considered in the study - work

remains to be done in the area of introducing domain

attributes and contextual information (along with in-

ternal attributes) as a basis for reusability metrics.

The authors of (Gui and Scott, 2009) propose new

coupling and cohesion metrics supporting the rank-

ing of software components extracted from the Inter-

net by a search engine, according to their reusabil-

ity level. Reusability is understood in terms of the

adaptation effort (NLOC added/deleted/modiﬁed) re-

quired to integrate a component into a larger system.

The novelty of their proposal lies in the fact that they

take into account both the transitive nature of cou-

pling/cohesion (indirect coupling/cohesion) and the

functional complexity of software components. The

performance of the newly proposed metrics in pre-

dicting reusability is evaluated using both linear re-

gression and Spearman rank correlation approaches,

on three software systems, each consisting of 20-25

components. In all cases, each of the new metrics

performs better than the existing four it is compared

against, showing the feasibility of the proposal.

In (Mija

c and Stapic, 2015), the goal is to conduct

a systematic literature survey in order to discover rel-

evant reusability metrics for software components. A

total of 39 papers have been investigated and summa-

rized, leading to the identiﬁcation of 36 quality factors

inﬂuencing reusability and 37 reusability metrics (12

for black box components and 25 for white/glass box

components).

The authors of (Ampatzoglou and Stamatia, 2018)

propose a new reusability predictor for software as-

sets, called Reusability Index (REI), that aggregates

various metrics corresponding to both structural and

external quality factors. REI is derived using back-

ward regression, taking into account 7 such factors

and associated metrics. The new measure is validated

against two existing proposals for assesing reusabil-

ity, using a dataset consisting of 15 projects extracted

from the Maven repository, both libraries and frame-

works. The ground-truth value is represented by the

number of times a given asset has been reused, as re-

ported by the Maven statistics.

Our study brings new elements to the body of

knowledge in this domain by: (i) conducting an in-

spection of reusability evolution in software systems,

by analysing how it changes in consecutive versions

of these systems; (ii) comparing different reusability

models by means of empirical investigation, namely

exploratory case study.

3 CASE STUDY DESIGN

3.1 Research Goals

Applying a Goal-Question-Metric approach (Basili

et al., 1994), the purpose of our study can be stated

as follows:

Analyze long term evolution of reusability

using three reusability computation models

for the purpose of evaluating them and their depen-

dencies to software metrics

from the source code point of view

in the context of several versions of large Java appli-

cations.

ICSOFT 2022 - 17th International Conference on Software Technologies

266

This general objective was then formulated into

research questions:

RQ1: How does reusability evolve over long term?

We investigate if and how the reusability indexes vary

during the evolution of the project, starting from early

versions of the applications.

RQ2: Which factors have an impact on reusability?

Different models use several metrics associated with

quality factors. We inquire which of these factors

makes a bigger impact on reusability. We call these

factors internal, since they are involved in the ana-

lyzed computational models of reusability.

RQ3: Which is the relation between reusability

and maintainability? Maintenance being one of the

most time consuming activities in software lifecycle,

the relation between reusability and maintainability

has been intensively studied (Lee and Chang, 2000;

Henry and Lattanzi, 1994). Even more, the cur-

rent software quality standard ISO 25010 (ISO/IEC

25010, 2011) considers reusability as a subcharacter-

istic of maintainability. As a consequence, we are in-

terested in how this relation is reﬂected in reusability

models.

The investigation has been designed as an ex-

ploratory case study, since our intention is to evaluate

the hypothesis formulated in the research questions

and has been developed using adopted standard in the

domain (Ralph, Paul (ed.), 2021) and similar prac-

tices described in literature (Kitchenham et al., 1995;

Runeson and H

ost, 2008).

3.2 Case study setup

To investigate the dependencies and differences be-

tween reusability models, we designed an exploratory

case study in which three such models were applied

to a set of three projects.

Selection of Reusability Estimation Models: We

have considered the following criteria when selecting

a model as part of our case study:

• provides a numerical assessment of reusability,

with clear calculation instructions;

• presents a solid proof of validity and has been

tested on a generous benchmark to conﬁrm its re-

liability;

• the metrics used are easily available and com-

putable by existing tools.

The candidate models, means of computation and

excluding criteria are presented in Table 1.

The ﬁnal selection contains: the model pro-

posed by the authors of (Papamichail et al., 2019) -

acronymed by us as PDS, the Taibi model introduced

in (Taibi, 2014), and QMOOD (Quality Model for

Object-Oriented Design) - presented in (Bansiya and

Davis, 2002).

The PDS model estimates reusability by aggre-

gating six quality factors (cohesion, coupling, com-

plexity, size, inheritance and documentation) and as-

sociated metrics, as shown in Table 2. For each of

the metrics, a polynomial regression model is trained,

translating its value into a reusability score. The

ground-truth value is taken as the number of occur-

rences within the import statements of a chosen set

of 3000 GitHub projects, extracted using an in-house

tool called Agora. Below, are the steps we followed in

implementing the PDS model, following the general

method proposed by the authors:

1. Setup Agora locally, then download the set of 100

projects used as benchmark and the 3000 projects used

for the ground-truth value;

2. For each of the 100 benchmark projects, compute all

static analysis metrics considered, at class level, using

SourceMeter;

3. For each class within each of the 100 benchmark

projects, retrieve its reuse rate, as the number of times

its name appears in the import statements of the 3000

projects dataset;

4. Eliminate the classes with 0 reuse rate (they are mostly

private and cannot provide relevant information);

5. Extract a general distribution for each metric, using a

binning strategy (a number of bins of ﬁxed size is cho-

sen, each containing all the classes having the value of

the metric within the bin boundaries);

6. Assign a reusability score to each bin, by summing up

the reuse rates of all classes belonging to the bin;

7. Normalize the scores from the previous step between 0

and 1;

8. Create the polynomial regression model corresponding

to each metric (that translates its value into a reusability

score), using pairs of data of the shape [BinCenter, As-

sociatedReusabilityScore]. The optimal degree for the

polynomial is obtained by using the elbow method on

the Root-Mean-Square-Error (RMSE);

9. Compute the reusability score corresponding to each

of the six quality factors, by aggregating the scores of

its associated metrics, using the formula (Papamichail

et al., 2019)

Score

f actor

∑

i=1

∗ Score

metric

∑

i=1

Above, N is the number of metrics used for the quality

factor considered, Score

metric

is the reusability score

predicted by the regression model for metric

, i = 1, N

and w

is the weight corresponding to metric

(com-

puted as the Pearson correlation between the values of

the metric and the associated reuse rates).

10. Compute the ﬁnal reusability score of the component,

by aggregating the scores corresponding to all factors

Empirical Evaluation of Reusability Models

267

Table 1: Synthesis of Reusability Estimation Models.

Paper describing the model Computation Excluding criteria

Reusability Index: A Measure for Assessing Software

Assets Reusability (Ampatzoglou and Stamatia, 2018)

Backwards Linear Re-

gression

Calculation only at

component and asset

level

Measuring the Reusability of Software Components

using Static Analysis Metrics and Reuse Rate Infor-

mation (Papamichail et al., 2019)

Hierarchical approach,

polynomial regression

accepted

A Reusability Evaluation Model for OO-Based Soft-

ware Components (Sandhu and Singh, 2008)

Neuro-fuzzy inference

engine

Only structural quality

factors considered

Estimation of Software Reusability for Component

based System using Soft Computing Techniques

(Singh et al., 2014)

Neural Networks Requires implementa-

tion of neural network

in Matlab

Empirical Analysis of the Reusability of Object-

Oriented Program Code in Open-Source Software

(Taibi, 2014)

Based on three quality

factors

accepted

A hierarchical model for object-oriented design qual-

ity assessment (Bansiya and Davis, 2002)

Hierarchical, based on

empirical and anecdo-

tal info

accepted

Estimation of Software Reusability: An Engineering

Approach (Nair and Selvarani, 2010)

empirical, weighted

combination of poly-

nomials

only structural code

quality characteristics

considered in the model, using the formula

ReusabilityScore =

∑

i=1

Score

f actor

The model proposed by Taibi (Taibi, 2014) is

based on three factors impacting the reusability of a

class: modularity (M), low complexity (LC) and un-

derstandability (U). As shown in Table 2, M is es-

timated based on the CBO and LCOM metrics, LC

based on ACC, DIT and NM, while U is computed by

aggregating the ROI and CIC metrics. The steps we

followed in order to estimate the reusability of a class

component are:

1. Run SourceMeter in order to obtain the values of

the ﬁve metrics quantifying M and LC for the

class (namely CBO/LCOM and ACC/DIT/NM,

respectively);

2. Apply a similarity-based comparison of class

names to code identiﬁers (for ROI) and comments

to identiﬁers (for CIC), using a pretrained NLP

model provided by the Gensim library (Gensim,

2021), in order to asses the values of the two met-

rics quantifying U;

3. Compute the reusability score F corresponding to

each of the three factors (see formula in (Taibi,

2014).

4. Compute the ﬁnal reusability score corresponding

to the class, based on the formula

R =

∑

i=1

∗ F

where F

and λ

, i = 1, n are the reusability scores

corresponding to the considered factors and the

tuning parameters, respectively.

Regarding the tuning parameters, a heuristic method

has been used, and extensive experiments have been

conducted, in order to determine the appropriate val-

ues. The best results have been reported for α

= 1.5

for CC, DIT and CBO, α

= 1 for NM and LCOM,

= 0.3 for U and λ

= 0.35 for M and LC (more

details can be found in (Taibi, 2014)).

QMOOD is a model used to assess high-level

quality factors of object-oriented systems: reusability,

ﬂexibility, understandability, functionality, extensibil-

ity and effectiveness. Each quality factor is mapped

to some design properties, whose quantiﬁcation relies

on appropriate metrics. In case of reusability, the cor-

responding design properties are coupling, cohesion,

design size and messaging, the four associated met-

rics used to estimate them being listed in Table 2. The

reusability score associated to a component is com-

puted using the formula (Bansiya and Davis, 2002)

Reusability = − 0.25 ∗Coupling+

0.25 ∗Cohesion

+ 0.5 ∗ Messaging

+ 0.5 ∗ DesignSize

Our own tool was developed implementing

the computational methods corresponding to these

models, while the metrics were computed using

SourceMeter (SourceMeter, 2022). All the charts

from next section have been generated with this tool.

ICSOFT 2022 - 17th International Conference on Software Technologies

268

Table 2: Overview of quality factors and metrics used in the analyzed models.

Factors Metrics Reusability models

Short

name

Meaning PDS Taibi QMOOD

Complexity

ACC Average Cyclomatic Complexity X

NL Nesting Level X

NLE Nesting Level Else-If X

WMC Weighted Methods per Class X

Coupling / Modularity

CBO Coupling Between Objects X X X

CBOI Coupling Between Objects Inverse X

NII Number of Incoming Invocations X

NOI Number of Outgoing Invocations X

RFC Response set For Class X

Cohesion LCOM5 Lack of Cohesion in Methods 5 X X X

Understandability

ROI Relevance of Identiﬁers X

CIC Correlation Identiﬁers Comments X

Documentation

AD API Documentation X

CD Comment Density X

CLOC Comment Lines of Code X

DLOC Documentation Lines of Code X

PDA Public Documented API X

Inheritance DIT Depth of Inheritance Tree X X

Size/Messaging

LOC Lines of Code X

LLOC Logical Lines of Code X

TNA Total Number of Attributes X

NOC Number of Children X

NM Number of Methods X

NPM Number of Public Methods X

NG Number of Getters X

TNOS Total Number of Statements X

Selection of Applications: The projects on which

we have applied the previously presented models are:

JUnit4 (JUnit4, 2021) - a widely adopted unit test-

ing framework for Java, based on annotations, Mock-

ito (Mockito, 2021) - a popular mocking framework

for unit tests in Java, and Atmosphere (Atmosphere,

2021) - a framework used for building asynchronous

Web applications. There are all mature systems, but

quite different in size, adoption and popularity lev-

els. Size and reputation indicators of these applica-

tions are summarized in Table 3.

Table 3: Details about size and reputation of applications.

Mockito JUnit4 Atmosph.

Lines of code 90,926 47,402 61,841

No of classes 1,970 1,505 899

Versions 605 28 219

Used by 107k 1.8m 134

GitHub stars 8.2k 12.1K 3.5

The research approach consisted in implementing

the three mentioned models of reusability applied to a

set of versions of the selected applications and inves-

tigating, on one hand, the differences in each applica-

tion, and on the other hand analyzing the differences

between the reusability estimation models. In order to

get a longitudinal view with respect to the evolution

of reusability throughout the lifetime of the selected

projects, we have applied the chosen reusability mod-

els on 23 versions of JUnit (out of 28), 45 versions of

Mockito (out of 605) and 30 versions of Atmosphere

(out of 219).

4 RESULTS

This section will offer answers to the formulated re-

search questions based on quantitative analysis of dif-

ferent computed metrics.

RQ1: How Does Reusability Evolve over Long Term?

In order to investigate the evolution of reusability

indexes throughout selected projects, we computed

the reusability indexes based on the three estimation

models, for the most signiﬁcant versions of each se-

Empirical Evaluation of Reusability Models

269

lected application: PDS shown in Figure 1, Taibi

in Figure 2, respectively QMOOD in Figure 3. We

perform a side-by-side comparison of the plotted

reusability curves produced by each model across the

release version history of the analyzed projects.

Analyzing the reusability evolution graphs, we

can observe that there is indeed some correlation be-

tween the curves predicted by the 3 models. In partic-

ular, we notice a common evolution pattern between

the values obtained with the PDS model and the ones

obtained with Taibi, for the Atmosphere and Mockito

projects. For JUnit4, the two models seem to perform

proportionally inverse. PDS values are on an ascend-

ing curve, while reusability reported by Taibi seems

to decrease sharply at the same point PDS is picking

up in value and continues to maintain an almost con-

stant value, with a very slight increase between the

following versions. QMOOD seems to yield the most

different results out of the 3 models.

We ﬁnd that there is very little correlation between

this model’s predicted values and the other two. For

the ﬁrst project, we can observe that QMOOD av-

erage values seem to be increasing by each version,

whereas PDS and Taibi have classiﬁed its reusabil-

ity as decreasing. We can, however, observe a slight

correlation between QMOOD and PDS where the JU-

nit4 project is concerned. Both models categorize

the project’s reusability level overall to be increasing,

though in the case of QMOOD the angle of the curve

is much sharper.

Lesson Learned: There are two conclusions that

we can draw from this exploration: Firstly, we

can say that the variation tends to appear in earlier

versions, and the later versions of the applications

become somewhat constant in terms of reusability

scores. Secondly, we can conclude that the PDS and

Taibi models perform similarly in terms of average

reusability evolution tracking, with QMOOD display-

ing very slight similarity to the other two.

RQ2: Which Factors Have an Impact on Reusability?

Table 2 summarizes the quality factors and associated

metrics used by the three computation models. We

ﬁrst remark the overwhelming number of metrics con-

sidered by PDS compared with the other two mod-

els. PDS also includes metrics associated with all

factors that have been proven to inﬂuence reusabil-

ity (Mehboob et al., 2021) (nominated in ﬁrst column

of the table). The Taibi model considers complexity,

coupling, cohesion, inheritance and size in its compu-

tation, while QMOOD is using only four metrics (as-

sociated to coupling, cohesion and size/messaging),

and ignores complexity and inheritance.

Then, we perform an in depth analysis in order to

investigate the relationships between constituent fac-

tors and reusability for different models. The tool al-

lows the selection of reusability score corresponding

to a model, respectively the factors: cohesion, com-

plexity coupling, documentation, inheritance or size.

The strongest inﬂuences detected are:

PDS - Complexity - the complexity attribute seems to

have the biggest impact on the ﬁnal reusability curves

observed for the PDS model. This is visible in Figure

4. We can observe a very strong correlation between

complexity and the ﬁnal reusability values for each of

the projects. However, in the particular case of Atmo-

sphere, the relationship between reusability and com-

plexity is an inverse one, which does not align with

the other two projects, for which the relationship is a

direct one. Intuitively, however, it seems more natural

to have reusability decreasing along with increasing

complexity, although this is not the case for the ﬁrst

project.

Taibi - Modularity - we observe a very strong direct

correlation between the evolution of modularity and

the one of reusability in the Taibi model (see Figure

5).

QMOOD - Messaging (quantiﬁed in the model as the

Number of Public Methods) - we distinguish a very

obvious inverse correlation between the average val-

ues of the Messaging attribute and the reusability ones

reported by the QMOOD model. Hence, we can say

that messaging seems to have a negative inﬂuence on

reusability (as also noted in Figure 6). We also ﬁnd

Cohesion to have a meaningful impact on the ﬁnal re-

sults obtained for QMOOD, but in a positive manner.

Lesson Learned: Our case studies conﬁrmed previ-

ous results from literature, in terms of the impact that

complexity, cohesion, coupling, size, inheritance and

documentation have on reusability. However, the met-

rics associated with these characteristics present dis-

tinct correlations to reusability.

RQ3: Which Is the Relation between Reusability and

Maintainability?

We have used Maintainability Index - MI (Oman

and Hagemeister, 1992) as the measure associated to

maintainability. The results of our evaluation are de-

picted in Figures 7, 8, respectively 9. The most promi-

nent correlation to maintainability is displayed by the

QMOOD model, which succeeds in tracking main-

tainability evolution trends for 2 out of the 3 projects

analyzed. The relationship seems to be a strong one

and is backed up by the design of the model itself,

which uses a number of overlapping metrics for both

reusability and maintainability, so a good estimation

of maintainability should also translate into a good es-

timation of reusability.

As far as the other two models are concerned, PDS

does not seem to correlate very much to maintainabil-

ICSOFT 2022 - 17th International Conference on Software Technologies

270

Figure 1: PDS Reusability Evolution (tool view with PDS model selected).

Figure 2: Taibi Reusability Evolution (tool view

with Taibi model selected).

Figure 3: QMOOD Reusability Evolution (tool view with

QMOOD model selected).

Figure 4: PDS Complexity impact on Reusability.

ity, however Taibi succeeds in returning close values

to the ones calculated for maintainability, although

their evolution curves seem to have little resemblance.

Lesson Learned: Considering the strong interdepen-

dency between reusability and maintainability, the re-

sults obtained in this case study are not very convinc-

ing. From our point of view, they in fact represent

another argument why maintainability index is not a

good indicator for maintainability (Molnar and Mo-

togna, 2021), (Heitlager et al., 2007), as it is depre-

cated: it was deﬁned in 1992, considering modular

and procedural programming languages, and ignor-

Empirical Evaluation of Reusability Models

271

Figure 5: Taibi Modularity impact on Reusability.

Figure 6: QMOOD Messaging impact on Reusability.

ing object oriented features that deﬁned new relations

such as inheritance, coupling and cohesion. However,

considering that MI is still used as indicator for refac-

toring by practitioners, and it is implemented in sev-

eral metric tools, we considered the investigation be-

tween MI and reusability to be relevant.

5 THREATS TO VALIDITY

The case study was conducted according to the exist-

ing standard (Ralph, Paul (ed.), 2021) and practices

(Runeson and H

ost, 2008). We started by formulating

the objective and research questions, then selecting

the reusability models and target applications. Data

collection and analysis was performed with the help

of the developed tool.

Internal threats were mitigated by using as much

as possible existing tools, already used in previous re-

search studies, and also by manually examining the

source code. However, the major threat is repre-

sented by the implementation of the reusability mod-

els, given the fact that the existing references have

not always been clear enough (such as Design Size

in QMOOD, Relevance of Identiﬁers and Correlation

Identiﬁers Comments in Taibi model). We address

this threat by looking for supplementary references

for these metrics in our implementation.

External threats are representing by the selection

of target applications, and we tried to address this

threat by considering open source projects, providing

access to source code and exposing a signiﬁcant trust

due to their intense use by the development commu-

nity. Section 3.2 contains details about the selection

process. Another threat might be represented by a

small amount of analyzed projects. This is one as-

pect that we consider for future improvement, but at

this stage of the project, we give a closer attention to

having a signiﬁcant number of versions for a project

rather than a large number of projects.

6 CONCLUSIONS AND FUTURE

WORK

We successfully implemented three different reusabil-

ity assessment models and applied them to a set of

open-source projects, that gave us the opportunity to

study reusability scores throughout signiﬁcant ver-

sions of selected projects, so that long term evolu-

tion of reusability can be observed. We also inves-

tigated the long-term relationship of various source

ICSOFT 2022 - 17th International Conference on Software Technologies

272

Figure 7: Relationship between reusability and maintainability - Atmosphere.

Figure 8: Relationship between reusability and maintainability - Mockito.

Figure 9: Relationship between reusability and maintainability - JUnit4.

code quality factors to measured reusability and found

the ones that have the biggest inﬂuence on the evo-

lution of the reusability degree. Finally, relation be-

tween reusability and maintainability was subject of

exploration.

Our ﬁndings are useful for both research and

practitioners’ communities. From a research point

of view, we bring empirical evidence in comparing

reusability models deﬁned in literature. For practi-

tioners, the developed tool can be integrated in or-

Empirical Evaluation of Reusability Models

273

der to track and control reusability scores, by setting

acceptable thresholds within projects. High quality

of code can be maintained in an automatic manner,

ensuring that no poorly written and low-reusability

components are introduced in the codebase – this is

especially useful for projects that serve as libraries

or frameworks, because the degree of reusability for

these must always be maintained high.

As future research directions, we consider this

study as a starting point that can be continued by:

• Adding more Reusability Assessment Models

to the performed analysis, including fuzzy, neu-

ral network-based ones and the whole spectrum of

machine learning models that has been proposed

in the literature as valid possible ways of assessing

reusability. This would be a great enhancement

to the current work and would deﬁnitely provide

promising results;

• Extending the Data Set of applications, in order

to increase the conﬁdence of the evaluation;

• Performing a more in Depth Analysis of

Reusability at the level of components and

classes.

ACKNOWLEDGEMENTS

This study was partially funded by the 2022 Develop-

ment fund of UBB.

REFERENCES

Ampatzoglou, A. and Stamatia, B. a. (2018). Reusability in-

dex: A measure for assessing software assets reusabil-

ity. In Capilla, R., Gallina, B., and Cetina, C., editors,

New Opportunities for Software Reuse, pages 43–58,

Cham. Springer International Publishing.

Atmosphere (2021). https://github.com/Atmosphere/

atmosphere. Online, accessed 01-03-2022.

Bansiya, J. and Davis, C. (2002). A hierarchical model

for object-oriented design quality assessment. IEEE

Transactions on Software Engineering, 28(1):4–17.

Basili, V., Caldiera, G., and Rombach, H. D. (1994). The

goal question metric approach. In Encyclopedia of

software engineering.

Frakes, W. and Terry, C. (1996). Software reuse: Metrics

and models. ACM Comput. Surv., 28:415–435.

Gensim (2021). https://radimrehurek.com/gensim/. Online,

accessed 01-03-2022.

Gui, G. and Scott, P. D. (2009). Measuring software com-

ponent reusability by coupling and cohesion metrics.

Journal of Computers, pages 797–805.

Heitlager, I., Kuipers, T., and Visser, J. (2007). A practi-

cal model for measuring maintainability. In Quality of

Information and Communications Technology, 6th In-

ternational Conference on the Quality of Information

and Communications Technology, QUATIC 2007, Lis-

bon, Portugal, September 12-14, 2007, Proceedings,

pages 30–39.

Henry, S. and Lattanzi, M. (1994). Measurement of soft-

ware maintainability and reusability in the object ori-

ented paradigm. http://hdl.handle.net/10919/19813.

ISO/IEC 25010 (2011). Systems and software engineering.

http://www.iso.org. Accessed: 2015.

JUnit4 (2021). https://junit.org/junit4/. Online, accessed

01-03-2022.

Kitchenham, B., Pickard, L., and Pﬂeeger, S. L. (1995).

Case studies for method and tool evaluation. IEEE

Softw., 12(4):52–62.

Lee, Y. and Chang, K. H. (2000). Reusability and main-

tainability metrics for object-oriented software. page

88–94. Association for Computing Machinery.

McCall, J., Richards, P., and Walters, G. (1977). Factors in

software quality. Nat Tech.Information Service, 1:0–0.

Mehboob, B., Chong, C. Y., Lee, S., and Lim, J. (2021).

Reusability affecting factors and software metrics for

reusability: A systematic literature review. Software:

Practice and Experience, 51.

Mija

c, M. and Stapic, Z. (2015). Reusability metrics of

software components: Survey.

Mockito (2021). https://site.mockito.org/. Online, accessed

01-03-2022.

Molnar, A.-J. and Motogna, S. (2021). A study of main-

tainability in evolving open-source software. In Ali,

R., Kaindl, H., and Maciaszek, L. A., editors, Eval-

uation of Novel Approaches to Software Engineering,

pages 261–282. Springer International Publishing.

Nair, T. G. and Selvarani, R. (2010). Estimation of soft-

ware reusability: An engineering approach. SIGSOFT

Softw. Eng. Notes, 35(1):1–6.

Oman, P. and Hagemeister, J. (1992). Metrics for as-

sessing a software system’s maintainability. In Pro-

ceedings Conference on Software Maintenance 1992,

pages 337–344.

Papamichail, M. D., Diamantopoulos, T., and Symeonidis,

A. L. (2019). Measuring the reusability of software

components using static analysis metrics and reuse

rate information. Journal of Systems and Software,

158:110423.

Poulin, J. S. (1994). Measuring software reusability. In

Proceedings of the Third International Conference on

Software Reuse: Advances in Software Reusability,

pages 126–138. Society Press.

Ralph, Paul (ed.) (2021). ACM Sigsoft Empirical Standards

for Software Engineering Research, version 0.2.0.

Runeson, P. and H

ost, M. (2008). Guidelines for conduct-

ing and reporting case study research in software en-

gineering. Empirical Software Engineering, 14:131–

164.

Sandhu, P. and Singh, H. (2008). A reusability evalua-

tion model for oo-based software components. World

ICSOFT 2022 - 17th International Conference on Software Technologies

274

Academy of Science, Engineering and Technology, In-

ternational Journal of Computer, Electrical, Automa-

tion, Control and Information Engineering, 2:912–

917.

Singh, C., Pratap, A., and Singhal, A. (2014). Estimation

of software reusability for component based system

using soft computing techniques. In 2014 5th Interna-

tional Conference - Conﬂuence The Next Generation

Information Technology Summit (Conﬂuence), pages

788–794.

SourceMeter (2022). https://www.sourcemeter.com/. On-

line, accessed 01-03-2022.

Taibi, F. (2014). Empirical analysis of the reusability

of object-oriented program code in open-source soft-

ware. International Journal of Computer and Infor-

mation Engineering, 8(1):118 – 124.

Empirical Evaluation of Reusability Models

275