SOFTWARE EFFORT ESTIMATION MODEL BASED

ON USE CASE SPECIFICATION

Xinguang Chen, Fengdi Shu

Lab for Internet Software Technologies, Institute of Software Chinese Academy of Sciences, Beijing, China

Ye Yang

Lab for Internet Software Technologies, Institute of Software Chinese Academy of Sciences, Beijing, China

Keywords: Use case, Software effort, Use case specification, Software estimation.

Abstract: Software effort estimation is essential for the project planning. Use case is widely used to capture and

describe the requirements of customers and used as an index of software measurement and estimation.

Based on the framework of traditional use case point estimation model, the paper presents UCSE, an effort

estimation model based on use case specification. Firstly, the model abstracts factors influencing software

effort from the use case specification and calculates the Use Case Weight, which is a kind of measurement

of use cases size. Secondly, a function is constructed to translate the software size expressed by Use Case

Weight to by software scale whose unit is kilo source line of code (KSLOC). Subsequently, effort

estimation model COCOMO II is used to estimate the software effort according to the estimated software

size measured by KSLOC. Compared with the traditional Use case point estimation model, UCSE model

makes use of more relevant information and is more operable since it provides more concrete and objective

references for the analysis and measurement of software effort factors in Use Case. What’s more, the

presented case study shows its results are more stable.

1 INTRODUCTION

How to select effective ways to improve accuracy of

software cost estimation has been the key in the

planning stage of software development. Long

software development cycle, many influencing

factors, emergencies and accidental events during

the software process lead to that people usually

make decision subjectively. That results in the

objectivity and accuracy in software cost estimation.

In 2004, Standish Organization statistics showed that

among more than 50,000 software projects, the ratio

of projects which could be completed is 29%, the

ratio of projects which could be questioned is 53%,

the ratio of projects which could be failed or

cancelled is 18%, and the ratio of projects which

could be finished but exceeded is 40% (Standish,

2004). It is widely believed that the main reason is

that people are lack of software effort estimation.

From the 1860s, software effort estimation

makes strong progress. At the beginning of research,

researchers have constructed the software effort

estimation model according to the characters of

software development and simple algorithm, such as

the SDC linear model (Boehm, 2005). Barry W.

Boehm put forward the COCOMO 81 (Boehm, 1981)

in 1981 and the COCOMO II (Boehm, 2000) in

2000. The COCOMO series establish the

relationship between software effort and software

scale, and adjust the function by a series of cost-

driving factors, which are the popular model.

As more and more software projects use the

unified modelling language (UML) to develop, use

case model is more and more used to capture and

describe the requirements of software. According to

the research of Neill in 2003, there are about 50%

projects adopting use cases or scenes to describe the

functional requirement (Neill, 2003). In 1993, the

first effort estimation method using use cases

(Karner, 1993) was proposed by Dr. Karner, which

was called use case point (UCP). Based on the UCP,

there were many related studies, which could be

generally classified to the 3 groups.

226

Chen X., Shu F. and Yang Y..

SOFTWARE EFFORT ESTIMATION MODEL BASED ON USE CASE SPECIFICATION.

DOI: 10.5220/0003463302260231

In Proceedings of the 6th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE-2011), pages 226-231

ISBN: 978-989-8425-57-7

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

Firstly, many case studies were used to validate

UCP. In 1999, John Smith from IBM proposed the

estimation framework based on the use case (Smith,

1999). The framework consists of five-level

structure which includes class, subsystem,

subsystem group, system, and multisystem.

Secondly, many researches were focused on the

improvements of UCP. In 2005, Carroll added a risk

parameter to the UCP (Carroll, 2005) to improve the

accuracy of UCP.

Thirdly, many researchers use UCP in different

specific areas. Nageswaran built the map from use

cases to test cases by using UCP and estimate the

test effort (Nageswaran, 2001).

To extract more concrete and objective

references for the analysis and measurement of

software effort factors in Use Case, the paper

proposes Use Case Specification Estimation model

(UCSE). The paper is organized as follows: section

2 discusses related and previous work; section 3

elaborates UCSE and section 4 presents a case study

of the model; and section 5 concludes the paper.

2 RELATED AND PREVIOUS

WORK

2.1 Use Case Point Model

Firstly, the model calculates weight of actor and

weight of use case to get the unadjusted use case

points (UUCP); secondly, the model adjusts UUCP

using technical factors and environment factors to

get UCP; finally, it establishes the relationship

between UCP and effort. Figure 1 shows the

procedure of UCP method.



Figure 1: Use case point model.

1）Unadjusted Actor Weight

The model classifies actors into 3 types: Simple,

Normal and Complex. We count the number of

actors in each type and multiply weighting factor

(WF) which the actor of every type corresponds to.

The unadjusted actor weight (UAW) is the sum of

all the results above. The symbol n is the quantity of

use cases.

UAW= Actor *WF

i=1

∑

(1)

2）Unadjusted Use Case Weight

The model classifies use cases into 3 types

according to the scale of use case: Simple, Normal

and Complex. We count the number of use cases in

each type and multiply weighting factor (WF) which

the use case of every type corresponds to. The

unadjusted use case weight (UUCW) is the sum of

all the results above. The symbol n is the quantity of

use cases.

UUCW= UseCase *WF

i=1

∑

(2)

3）Unadjusted Use Case Points

UAW plus UUCW equals unadjusted use case points

(UUCP).

UUCP=UAW+UUCW

(3)

4）Other factors

The model uses technical complexity factors (TCF)

and environmental complexity factors (ECF) to

adjust the UUCP. TCF reflects the internal

properties of software, such as security, reusability

etc. ECF reflects the external properties of software,

such as personnel experience etc. Every factor has a

value of 0 to 5. The value reflects the degree of

every factor and judged subjectively. The equations

of TCF and ECF are shown below:

i=1

TCF=0.6+0.01*( T *Weight )

∑

(4)

i=1

ECF=1.4+(-0.03)*( E *Weight )

∑

(5)

We can calculate the UCP and Effort using

Equation (6) and (7):

UCP=UUCP*TCF*ECF

(6)

Effort=UCP*Productivity Factor

(7)

Productivity Factor is a coefficient of software

project which is calculated by historical data. Dr.

Karner thought that productivity factor equalled to

20 man-hours.

The traditional UCP considered sufficiently

information of use case in software development and

some impacts, but there are three problems: Firstly,

there is no definite reference to classify the actor and

use case. It is difficult to use. Secondly, some

SOFTWARE EFFORT ESTIMATION MODEL BASED ON USE CASE SPECIFICATION

227

information in use case which is also relevant with

software effort, such as extended event, is neglected.

Thirdly, the value of technical and environmental

factors is subjective.

2.2 COCOMO II

COCOMO II was proposed by Barry Boehm in 2000.

The model adjusts and updates the cost-driving

factors. EM is a multiplier of software effort; SF is

an index of scale factor; A and B are the parameters

whose value is adjusted by historical data.

Effort=A*(KSLOC) * EM

i=1

∏

(8)

E=B+0.01* SF

j=1

∑

(9)

To COCOMO II：A＝2.94, B=0.91

Every cost-driving factor is corresponding to a

value of EM. Every value of EM should be put into

the function above to calculate software effort, cost

and schedule. The key of COCOMO II is the Kilo

source line of code (KSLOC). The accuracy of effort

depends largely on the accuracy of software scale.

Boehm thought the important input in COCOMO II

is KSLOC (Boehm, 2000).

3 UCSE MODEL

UCSE consists of several parts. First, based on the

framework of the traditional UCP, we analyse the

structure of use case specification, calculate the use

case weight (UCW) according to the document of

use case specification, and establish the relationship

between UCW and KSLOC. Subsequently, software

effort is calculated by COCOMO II based on the

estimated size in KSLOC. Use case specification is

used because it has full information to make up the

second problem of UCP presented in section 2.1.

The overview of UCSE model is shown in

Figure 2 and its main components are introduced in

following.



Figure 2: UCSE Model.

3.1 Unadjusted Actor Weight

The type of actors is important. In the use case

specification, actors could be easy to classify.

According to UCP and use case specification, the

actor factor is described in Table 1, which including

its classes and corresponding classification rules and

weights.

Table 1: Actor Factor.

Actor Class Rule Weight

Simple Only one actor 1

Medium Two actors 2

Complex Three or more actors 3

As shown in formula 10, unadjusted actor weight

(UAW) equals to the sum of the product of actor

class (AC) and class weight (CW), where n is the

quantity of use cases.

UAW = AC *CW

i=1

∑

(10)

3.2 Event Weight

Event flow in use case specification describes the

scene of use case and reflects the steps of software

execution. Event flow is the main part in use case

specification. It has two types: primary event and

extended event. Effort is influenced differently by

two types. Primary event flow is important to scale

of use case, while extended event flow occurs in

some abnormal or special occasions. The traditional

UCP neglected the extended event. That may

produce some errors.

UCSE model provides classification rules of the

primary event and each class of primary event has its

weight, which is shown in Table 2:

Table 2: Primary Event Factor.

Event Type Rule Weight

Simple

The number of primary event

flows ≤ 3

Medium

The number of primary event

flows is between 4 and 7

Complex

The number of primary events

flows ≥ 7

As shown in formula 11, unadjusted primary

event weight (UPEW) equals to the sum of the

product of primary event class (PEC) and class

weight (CW), where n is the quantity of use cases.

UPEW = PEC *CW

i=1

∑

(11)

ENASE 2011 - 6th International Conference on Evaluation of Novel Software Approaches to Software Engineering

228

UCSE model provides classification rules of the

extended event and each class of extended event has

its weight, which is shown in Table 3:

Table 3: Extended Event Factor.

Event Type Rule Weight

Simple

The number of extended event

flows ≤ 5

Complex

The number of extended event

flows ≥ 5

As shown in formula 12, unadjusted extended

event weight (UEEW) equals to the sum of the

product of extended event class (EEC) and class

weight (CW), where n is the quantity of use cases.

UEEW = EEC *CW

i=1

∑

(12)

As shown in formula 13, unadjusted event

weight (UEW) equals to the sum of UPEW and

UEEW.

UEW=UPEW+UEEW

(13)

3.3 Unadjusted Use Case Weight

Actor and event are the most important parts in use

case specification, so unadjusted use case weight

(UUCW) the sum of UAW and UEW.

UUCW=UAW+UEW

(14)

3.4 Business Rules

Traditional UCP has 13 TCFs and 8 ECFs, which

reflect two points in the software development: one

is restraint of actors; the other is the restraint of

operation rules. In the UCP, we identify these values

of factors subjectively. Business rules (BR) in the

document of use case specification reflect directly

the status of persons and technology and is more

objective to reflect the real status in software

development.

BR has two types: one is the global rules, the

other is special rules. Global rules usually are related

to every use case. For example, an actor needs to be

authorized to execute a use case, so the execution of

use case is corresponding to the level of Authority

Classes, or operations of users need to be recorded

in the system and so on. Special rule only exists in

the extended event. It cannot change as external

environment changes. For example, an order at least

has one type of goods, and the quantities of goods

need to be less than 5 and so on. Besides, data

verification belongs to special rule. For example, the

number of identification card must be 15 or 18

digits, and zip code must be 6 digits and so on. We

classify special rules into Standard, Medium and

Complex, which is shown in Table 4:

Table 4: Business Rules.

Type Rule Weight

Global Rules global rules exist 1.05

Special

Rules

Simple No special rule 1.0

Medium

≤ 50 percent of use

cases has special rules

1.5

Complex

≥ 50 percent of use

cases has special rules

Use case weight (UCW) equals to the product of

UUCW and BR.

UCW=UUCW*BR

(15)

3.5 Relationship between UCW and

KSLOC

KSLOC measures the scale of software, and UCW

measures the scale of use case. The UCSE model

establishes the relationship between KSLOC and

UCW.

Regression is a common statistical method to

determine the quantitative relationship between two

or more variables. Regression has two types: one is

linear regression, the other is non-linear regression.

Exponential function and logarithmic function are

common non-linear function. We will determine the

function between KSLOC and UCW by using data

of 14 projects. The function is shown as:

KSLOC=f( UCW)

(16)

3.6 Effort Function

We input KSLOC into COCOMO II and the

function of effort is shown as:

i=1

Effort=A*(f(UCW)) * EM

∏

(17)

j=1

E=B+0.01* SF

∑

(18)

4 CASE STUDY

We carry out a case study on 14 software projects

based on their use case specification. The first 10

projects (EP1 to EP10) are used to make regression

SOFTWARE EFFORT ESTIMATION MODEL BASED ON USE CASE SPECIFICATION

229

analysis on KSLOC and UCW. The other four (VP1

to VP4) are used to validate the UCSE model.

EP1 is a human resource system; EP2 is a tool

which integrating some estimation tools; EP3~EP10

are different versions of a software process

management platform. The data of these projects are

shown Table 5:

Table 5: Data of Projects.

Project Number of Use case UCW KSLOC

EP1 5 125 6.034

EP2 7 199.5 13.52

EP3 60 3307.5 169.7

EP4 164 4192 250

EP5 186 4110 194

EP6 189 3687.08 191

EP7 195 5471.55 228

EP8 222 6575.1 294.8

EP9 312 13657.8 618

EP10 278 16989 824.8

4.1 Regression Analysis

We use UCW and KSLOC to make regression

analysis. The result is shown in Figure 3:

Figure 3: Regression Curve.

The result of the three regression equations is

shown in Table 6:

Table 6: Regression Equation.

Regression Equation

Linear regression

0.046 13.43yx

0.988

Indicial Regression

0.001

41.31

ye=

0.595

Power Regression

0.962

0.069yx=

0.993

The result shows that the power function

regression equation is very simple and ideal. The

coefficient is highest (0.993). Therefore, UCSE

model uses the power function. The graph of power

function regression equation is shown in Figure 4:

Figure 4: Graph of power regression equation.

So the equation is shown as:

0.962

KSLOC=0.069*UCW

(19)

4.2 Result Analysis

VP1-VP4 was developed based on their use case

specifications. VP2 is a customized software process

management platform. The four projects were

estimated by UCP and UCSE model respectively to

get estimated effort. We calculate relative error,

average relative error and standard deviation to

compare two models. In Table 7, EE is the

Estimated Effort and RE is the Relative Error. The

result is shown in Table 7:

Table 7: Estimation Results of UCSE and UCP.

Real

Effort

UCSE Model UCP Model

EE RE EE EE

VP1 2281.39 2840.24 0.24 3114.40 0.37

VP2 3426.65 3865.64 0.13 3931.20 0.15

VP3 2625.58 2505.10 -0.05 2479.80 -0.06

VP4 3110.75 3697.62 0.19 3587.90 0.15

Average Relative

Error

0.1519 0.1803

Standard

Deviation

0.0739 0.1135

From Table 7, we can see that the estimation

results of UCSE model are better than those of UCP

model except as for VP4. The average relative error

and standard deviation of UCSE model are less than

the UCP model. The UCSE model analyses the

document of use case specification sufficiently,

includes the extended event which was neglected by

UCP model, improves the way to identify the weight

of actors and event flows, and uses business rules

instead of TCFs and ECFs. UCSE model improves

ENASE 2011 - 6th International Conference on Evaluation of Novel Software Approaches to Software Engineering

230

the operability so that the results are more accuracy

and steady.

5 CONCLUSIONS

Use case is a common tool and penetrates through

the software development. Use case specification is

the document of use case and is easy to reflect and

describe the software process. The paper proposes

UCSE model to improve the accuracy of estimating

and be more operable since it provides more

concrete and objective references for the analysis

and measurement of software effort factors in use

case. The result of UCSE model is better than

traditional model. However, the usage of the model

is related to the form of the use case specification,

which lacks uniform standard. So the future work

includes how to overcome this limit and enhance the

applicability of UCSE model.

ACKNOWLEDGEMENTS

This work is supported by the National Natural

Science Foundation of China under Grant

Nos.60873072

， 61073044, and 60903050; the

National Science and Technology Major Project; the

National Basic Research Program under Grant

No.2007CB310802; CAS Innovation Program.

REFERENCES

The Standish Group. (2004). The 3rd Quarter Research

Report. (n.d.), from http://www.standishgroup.com.

Boehm B. W., Valerdi R. (2005). Achievements and

challenges in software resource estimation. Retrieved

May 13, 2005, from http://sunset.usc.edu/public

ations/TECHRPTS/2005/usccse2005-513/usccse2005-

513.pdf

Boehm. B. W. (1981). Software Engineering Economics

ed.). NJ: Prentice-Hall

Boehm. B. W., Madachy R. and Steece B. (2000).

Software Cost Estimation with COCOMOII (1

ed).

NJ: Prentice-Hall

Guarv Karner. (1993). Resource Estimation for Objectory

Projects. Objective Systems SF AB (Copyright owned

by Rational Software)

John Smith. (1999). The estimation of effort based on use

cases. Rational Software White Paper

Neill C. J., Laplante P. A. (2003). Requirements

engineering: the state of the practice. IEEE. 20 (11).

40-45, doi:dx.doi.org/10.1109/MS.2003.1241365.

Carroll E. R. (2005). Estimating software based on use

case points. Proceedings of the Companion to the 20th

annual ACM SIGPLAN conference. 257-265.

doi:dx.doi.org/10.1145/1094855.1094960.

Nageswaran S. (2001). Test effort estimation using use

case points. Proceedings of the 14th Internet &

Software Quality Week.

SOFTWARE EFFORT ESTIMATION MODEL BASED ON USE CASE SPECIFICATION

231