Available online at www.sciencedirect.

com

Applied Soft Computing 8 (2008) 1266–1273

www.elsevier.com/locate/asoc

Fuzzy regression model of R&D project evaluation

Shinji Imoto a, Yoshiyuki Yabuuchi b, Junzo Watada a,*
a
Waseda University, 2-7 Hibikino, Wakamatsu-ku, Kitakyushu, Fukuoka 808-0135, Japan
b
Shimonoseki City University, 2-1-1 Daigaku, Shimonoseki, Yamaguchi 751-8510, Japan
Received 21 February 2007; accepted 23 February 2007
Available online 22 November 2007

Abstract
Engineering and technology play an important role in strengthening the competitive power of a company and in surviving a severe competition
in the world. About 70% of the total R&D investment in Japan comes from the private sector. It is the most important to decide which research
projects have to be adopted for a future research out of proposals from divisions and sections in a company. The objective of this paper is to analyze
the results of experts’ evaluation in selecting submitted proposals for R&D and to model the experts’ evaluation. This paper analyzes a research and
development of a certain manufacturing company in a heavy metallurgy industry.
We employed a principal component model, dual scaling, AHP and fuzzy regression analysis to analyze the results that experts evaluated
proposed research projects for single or plural of fiscal years. The experts’ evaluation was pursued on the basis of (1) the objective of a research
project, (2) its background, (3) its research contents, (4) the expected effect, (5) the possibility of obtaining patents, (6) project schedule, (7)
developing cost, etc. The obtained model results in the same selection of projects as the experts did.
# 2007 Elsevier B.V. All rights reserved.

Keywords: Management of technology and engineering; R&D; Project management; AHP; Fuzzy regression model

1. Introduction evaluated only by customers themselves, but also assessed in

various markets indirectly. Every company should not only
In an economically high-growth period of Japan after the realize the efficiency and strength of management, but also grasp
Second World War, companies competed with each other to the market change and movement under the consideration that all
manufacture and market products more efficiently and effectively companies are sorted through the severe evaluation in the market.
than another companies. Therefore, the companies in Japan This forces every company to continue and accelerate the
employed production control, new manufacturing methods, acquisition of new technology and knowledge. Therefore, the
quality control and so on in their factories. Various excellent research and development play an important role in a company. If
methods of production are created and developed such as just-in- the result of R&D leads a company to acquiring very profitable
time, automation, smoothening of production, etc. as a result. It is technology, the company would effectively utilize patents as an
also necessary to reengineer a business process. In recent years, a intellectual property. Of course, judging from an efficiency of
company is required to provide products according to ‘‘needs’’ fund management, an R&D expenditure and a royalty frequently
and ‘‘wants’’ of a market. Therefore, it is essential for a business forces us to accept the situation where they should abandon their
to grasp the ‘‘needs’’ and ‘‘wants’’ of a market. It can run a own research and development.
business to cultivate products and services according to them. Here, we have to pay attention to the application of intellectual
Generally, a company is evaluated based on the knowledge it property rights such as patents [1,2]. We focus on one company
possesses, and the creation and application of knowledge should that endeavors to get the basic and inclusive patents on the
be changed into a business. The company is not directly business basis through early technical development. Presently,
R&D is actively and widely executed with industry–university–
government cooperation propulsion business and technology
promotions policy, etc. It is strongly desired to enforce R&D
* Corresponding author at: Waseda University, IPS, 2-7 Hibikino, Waka-
sufficiently as competitiveness and to get excellent results. On
matsu, Kitakyushu, Fukuoka 808-0135, Japan. Tel.: +81 90 3464 4929;
fax: +81 93 692 5179. the other hand, a company realizes the effective and efficient
E-mail address: [email protected] (J. Watada). management of research and development.

1568-4946/$ – see front matter # 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.asoc.2007.02.024
 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273 1267

The objective of this paper is to understand the evaluation Table 2

The evaluation indices of R&D projects
structure of R&D projects at one comprehensive production
company. In this paper, we analyze which projects proposed No. The evaluate indices Note
should be really selected, and find the structure how to evaluate 1 Objective of R&D Fitness of business objective
R&D projects. 2 Necessity Necessity of the R&D
3 Technological difficulty Technological difficulty
2. R&D project 4 Period Required period for R&D
5 Cost Research fund
6 Expected sales volume The growing volume received order
The survival of a company depends on the acceleration of its 7 Possibility to get a patent Possibility to get a patent
technical development and the realization of the creative R&D
which can start one new business. Therefore, R&D is a driving
force of a company, and it is not only sufficient to build up the Index 1: ‘‘Objective of R&D’’ is to evaluate the objective
firm foundation of R&D in order to make the company expand fitting to a company’s mission, and projects included
but also to pursue wide activities in order to realize technical in each category are emphasized by the area pursued
development. by the company and set a high valuation.
One of the important strategies for a company is to make Index 2: ‘‘Necessity’’ is to evaluate the necessity to start a
clear the way to execute R&D, the rule to select R&D project.
projects and set priority levels to projects, etc. It depends Index 3: ‘‘Technological difficulty’’ is to evaluate the techno-
on the knowledge of an administrator and the thinking logical difficulty that a project faces.
of the executive in most cases. The objective of this paper is Index 4: ‘‘Period’’ is to evaluate the duration required until
to make the rule of evaluating R&D projects clear by obtaining some results of a project since the start of
analyzing data obtained from a company and to effectively the project.
select R&D projects that are required to achieve the Index 5: ‘‘Cost’’ is to evaluate the research fund required.
company’s goal. Index 6: ‘‘Expected sales volume’’ is to indicate how many
achievements will be obtained, that means the
2.1. R&D project and evaluation indices evaluation of the growing volume received orders
in the future. In other words, this means the project
Table 1 shows eighteen R&D projects proposed in the can create a product which will be sold largely.
company in 1998 that are scrutinized and studied in this Index 7: ‘‘Possibility to get patents’’ is to evaluate the
paper. Table 2 illustrates indices that are used to evaluate the possibility to get patents.
R&D projects. Each index is assessed into five levels of
ranking. These indices are evaluated using five levels of ranking.
Table 3 shows the evaluation results of these indices.
Table 1
R&D project in 1998 2.2. Evaluation of R&D project
No. R&D project
1 Development of electronic furnace control model based on fuzzy In the company concerned, the proposal of each R&D
neural network project is evaluated on the basis of the evaluation indices shown
2 Prototype development of 3 layer client server system in Tables 2 and 3. R&D projects get their priority according to
3 Establishment of a foundation technology of Windows NT server the evaluation. Then, R&D projects are selected within the
on internet
budget. Table 4 shows the total value, that is, the sum of
4 Development of new MINBAT control system
5 Development of an application system of image processing evaluated values of the indices. Therefore, an R&D project is
6 Commercialization of checking device for coming off a character highly evaluated and has a high priority, when it has a large total
string value.
7 Development of application software for Windows NT In this paper, this evaluating method is named ‘‘a
8 Development of intelligent Ethernet module for IBM PC/AT
conventional method’’ in order to distinguish from AHP and
compatibles
9 Development of reduction system of the peak electricity demand (1) principal component models in Section 4.
10 Development of a charge-discharge electricity system for an
electric car (2) 3. Feature extraction by principal component analysis
11 Development of transportable cycle tester
12 Development of an optical cutting profile meter
In this section, Table 4 shows R&D projects analyzed by a
13 Development of new automatic checker
14 Development of X-ray fluoroscope principal component model. Since Index 1 has the same value
15 Development of low price thickness measure for a small size and of Index 3 for all projects in Table 4, Index 1 is removed from
a thin board the analysis.
16 Development of X-ray CT system (1st step) Tables 5 and 6 show correlation coefficient obtained and
17 Commercialization of low cost X-ray CT system
factor loading by the principal component analysis, respec-
18 Development of application software for an account
tively. Since the accumulated proportion is 68.7% up to the 2nd
1268 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273

Table 3
Evaluation of importance degree of each R&D project
R&D project Necessity Technological Period Cost Expected sales Possibility to
difficulty volume get a patent
Developing device 5 Extremely 5 Extremely 1 Extremely long 1 More than 20 1 Extremely 5 Extremely 5
(longer than 1 year) million yen
Developing control 4 Extremely 5 Extremely 1 Extremely long 1 More than 20 1 Extremely 5 Extremely 5
model (longer than 1 year) million yen
Commercializing 4 Rather 4 Very 2 Very long (longer 2 More than 10 2 Very 4 Very 4
control model than 8 months) million yen
Developing and 4 Rather 4 Very 2 Very long (longer 2 More than 10 2 Very 4 Very 4
commercializing than 8 months) million yen
Developing system 3 Yes 3 Yes 3 Normal (longer 3 More than 5 3 Normal 3 Yes 3
than 6 months) million yen
Constructing system 3 Yes 3 Yes 3 Normal (longer 3 More than 5 3 Normal 3 Yes 3
than 6 months) million yen
Constructing a server 2 A little 2 Not so 4 Not so long (longer 4 More than 3 4 Not so 2 Not so 2
computer than 3 months) million yen
Developing software 2 A little 2 Not so 4 Not so long (longer 4 More than 3 4 Not so 2 Not so 2
than 3 months) million yen
Adding function 1 Not so 1 No 5 Short (shorter 5 Less than 3 5 No 1 No 1
than 3 months) million yen
Decided on function 1 Not so 1 No 5 Short (shorter 5 Less than 3 5 No 1 No 1
than 3 months) million yen
Acquire 1 No 0 No 5 Short (shorter 5 Less than 3 5 No 1 No 1
than 3 months) million yen

principal component shown in Table 6, it is sufficient to employ can be interpreted as a necessity of the R&D project in the
only the 1st and 2nd principal components for discussion in company.
Tables 6 and 7. Table 8 depicts sample scores. Next, on the 2nd principal component, the factor loading of
Fig. 1 illustrates category scores of both the 1st and 2nd Index 4 is 0.822, and Index 5 is 0.695. The correlation
principal components. Indices 2 ‘‘Necessity,’’ 3 ‘‘Technological coefficient of these indices is 0.254. The negative direction of
difficulty,’’ 6 ‘‘Expected sales volume’’ and 7 ‘‘Possibility to the 2nd principal component shows the shortness of R&D
get patents’’ are placed in the large negative value of the 1st duration and the positive direction shows small expense. In
principal component. These indices have a positive correlation. other words, it moves to negative direction of 2nd principal
Especially, Indices 2 and 3 have a strong correlation. component when R&D period is short, to positive direction
Considering the above mentioned, the 1st principal component when R&D expenditure becomes large. Therefore, R&D
project has a short developing period and small cost distribute
in the 3rd quadrant and the 4th quadrant. The R&D projects
Table 4
should have a large risk for the company in the case where the
Evaluation of R&D project
sample score on the 2nd axis becomes a large negative value.
Project no. Evaluate indices Total When we verify sample scores in Table 8, Project 17
1 2 3 4 5 6 7 ‘‘commercialization of a low cost X-ray CT system,’’ Project 16
1 4 5 4 1 3 5 5 27 ‘‘development of a medical X-ray CT system (1st step),’’
2 3 3 3 2 4 5 1 21 Project 12 ‘‘development of an optical cutting profile meter’’
3 2 2 2 2 3 3 1 15 and Project 13 ‘‘development of an X-ray Fluoroscope’’ have a
4 1 2 1 1 3 3 1 12 large negative value on the 2nd principal component. In other
5 4 4 4 1 3 3 5 24
words, it can be understood that the cost of development
6 5 4 5 1 3 4 3 25
7 1 2 1 1 1 3 1 10 becomes expensive in the negative direction of the 2nd
8 1 1 1 1 4 2 1 11 principal component.
9 1 1 1 1 1 2 1 8 Simply, it becomes as follows.
10 1 2 1 1 3 4 1 13 Fig. 2 illustrates sample scores shown in Table 8. Fig. 2
11 3 3 3 1 1 5 1 17
shows that the R&D project that has a large negative value on
12 1 2 1 3 1 2 1 11
13 4 4 4 1 2 5 1 21 the 1st principal component has a large evaluated value.
14 5 4 5 2 1 3 1 21 Summarizing these results when, in Fig. 2, R&D projects are
15 4 3 4 2 3 5 1 22 in the 2nd quadrant, and sample scores on the 1st principal
16 5 4 5 2 1 2 1 20 component are a large negative value and sample scores on the
17 5 4 5 3 2 2 1 22
2nd principal component are a large positive value. The project
18 1 2 1 1 4 2 1 12
should have a large evaluation.
 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273 1269

Table 5
Correlation coefficient between evaluation indices
Necessity Technological Period Cost Expected Possibility to
difficulty sales volume get a patent
Necessity 1 0.903 0.141 0.097 0.429 0.562
Technological difficulty 0.903 1 0.271 0.147 0.310 0.356
Period 0.141 0.271 1 0.254 0.269 0.311
Cost 0.097 0.147 0.254 1 0.194 0.235
Expected sales volume 0.429 0.310 0.269 0.194 1 0.237
Possibility to get a patent 0.562 0.356 0.311 0.235 0.237 1

Table 6
Category score
1st PC 2nd PC 3rd PC 4th PC 5th PC 6th PC
Necessity 0.959 0.201 0.034 0.026 0.066 0.182
Technological difficulty 0.871 0.376 0.035 0.070 0.275 0.139
Period 0.002 0.822 0.254 0.427 0.277 0.010
Cost 0.034 0.695 0.429 0.564 0.115 0.014
Expected sales volume 0.577 0.383 0.604 0.337 0.202 0.029
Possibility to get a patent 0.676 0.404 0.401 0.369 0.282 0.056
Eigenvalue 2.470 1.651 0.776 0.756 0.290 0.057
Proportion 41.167 27.518 12.933 12.600 4.837 0.945
Accumulated proportion (%) 41.167 68.685 81.617 94.218 99.055 100.000
PC: principal component.

Even if sample scores on the 2nd principal component have a 0.181. It is clarified that four evaluation indices have the similar
large negative value and an R&D project had great risk, the weight and the principal component analysis obtained the
R&D project has sufficient necessity to be selected. The project similar result. In addition, Indices 3 (technological difficulty)
should be evaluated at a large value. and 5 (cost) have the same value 0.100 and Index 4 (period) has
0.074. Table 4 shows the weights of both Indices 1 (objective of
4. Evaluation of R&D project by AHP R&D project proposal) and 3 (technological difficulty) have the
same value as the pair-wise comparison of AHP between both
Here, R&D projects are analyzed through AHP [3–5]. Tables indices are in Table 10.
9 and 10 illustrate the weights of evaluation indices and of the The evaluation score of an R&D project is obtained as shown
alternatives, respectively. Table 9 shows weights of Indices 1 in Table 11 by the calculation using Tables 9 and 10. The
(objective of R&D), 2 (necessity of R&D), 6 (Expected sales evaluation can be pursued by the method shown in Table 4. In
volume) and 7 (possibility to get patent) are the same value, other words, a score of conventional method is defined in

Fig. 1. Principal component analysis (scatter chart of factor loading).

1270 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273

Fig. 2. Principal component analysis (category scores).

Table 3, and is obtained by the sum total of evaluation indices. projects by principal component analysis and total score by
Concerning principal component analysis as mentioned above, AHP are shown in Table 11.
a project is highly evaluated when it has a large negative value In Table 11, the conventional evaluation method shows the
on the 1st principal component and a large positive value on the method to evaluate an R&D project by the sum total of index
2nd principal component. The total score by conventional score and to define a ranking by the total sum scores. From the
evaluation method and the sample score, ranking of R&D results of AHP and conventional evaluation method, we can
understand the difference of ranking among R&D Projects 2, 4,
12, 13, 14, 15, 16 and 17. Since there are R&D projects with the
Table 7 same score and these projects have a small difference among
Meaning of principal components them, there is no big difference in the results between both
Principal component Direction Meaning methods. Concerning R&D Project 16. There is a difference
1st Negative Necessity such that the ranking of AHP is 6th and conventional evaluation
2nd Positive Development cost method is 9th. In addition, on comparison of the evaluation with
principal component analysis, it seems the difference of
ranking is not big enough, but some difference is recognized.
Table 8
Sample score 5. Construction of R&D evaluation model
Project 1st PC 2nd PC 3rd PC 4th PC 5th PC 6th PC
5.1. Fuzzy regression analysis
1 3.177 1.513 0.412 0.584 0.827 0.266
2 0.472 0.552 0.223 2.021 0.298 0.000
3 1.028 0.116 0.423 0.769 0.283 0.161 A possibilistic regression model is built in terms of the
4 1.363 0.994 0.021 0.006 0.154 0.257 possibility and interpretes all observed values as possibilities
5 2.035 1.154 1.519 1.204 0.309 0.197 which the system should contain. Then, the possibilistic
6 2.016 0.740 0.273 0.180 0.494 0.293
regression model is also renamed as a possibilistic regression
7 1.403 0.021 0.896 1.160 0.230 0.150
8 2.180 1.369 0.953 0.291 0.552 0.254 model [6,7]. In other words, the possibilistic regression
9 2.239 0.091 0.359 1.458 0.025 0.414 model aims to build a model so that it could contain all
10 1.058 1.241 0.592 0.328 0.158 0.155 observed data in the estimated possibilistic numbers resulted
11 0.415 0.024 1.954 0.442 0.126 0.098 from the model.
12 1.713 2.090 0.514 0.054 1.413 0.127
13 1.303 0.196 1.459 0.164 0.483 0.267 Table 9
14 1.007 1.895 0.312 0.301 0.479 0.001 Weights of evaluation indices in AHP
15 0.789 0.112 0.636 1.487 0.179 0.408
16 0.701 2.143 0.258 0.624 0.791 0.101 Evaluation indices
17 0.719 2.588 1.115 0.674 0.236 0.092 1 2 3 4 5 6 7
18 1.649 1.232 0.986 0.266 0.659 0.412
Weight 0.181 0.181 0.100 0.074 0.100 0.181 0.181
PC: principal component.
 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273 1271

Table 10 The regression model with fuzzy coefficients can be

Weight of projects in AHP
described using lower boundary axj  cjxjj center axj and
R&D project Evaluation indices upper boundary axj + cjxjj. When a sample (yj, xj)(j = 1, 2, . . .,
1 2 3 4 5 6 7 m) with center yj and width dj is given as fuzzy number (yj, dj),
the membership function of fuzzy coefficients is set to L(u), and
1 0.073 0.125 0.073 0.032 0.060 0.101 0.176
2 0.046 0.046 0.046 0.061 0.102 0.101 0.034
possibility to which extent a sample is included in the
3 0.029 0.027 0.029 0.061 0.060 0.037 0.034 regression model is a, the inclusion relation between the model
4 0.020 0.027 0.020 0.032 0.060 0.037 0.034 and the samples should be written as follows:
5 0.073 0.079 0.073 0.032 0.060 0.037 0.161
6 0.116 0.079 0.116 0.032 0.060 0.063 0.093 y j þ L1 ðaÞd j  ax j þ L1 ðaÞcjx j j
(3)
7 0.019 0.027 0.019 0.032 0.021 0.037 0.034 y j  L1 ðaÞd j  ax j  L1 ðaÞcjx j j
8 0.019 0.017 0.019 0.032 0.102 0.022 0.034
9 0.019 0.017 0.019 0.032 0.021 0.022 0.034 In other words, the possibilistic regression model is built to
10 0.019 0.027 0.019 0.032 0.060 0.055 0.034
contain all samples in the model. This problem results in a
11 0.046 0.046 0.046 0.032 0.021 0.101 0.034
12 0.019 0.027 0.019 0.103 0.021 0.024 0.034 linear programming.
13 0.073 0.079 0.073 0.032 0.035 0.101 0.034 Using the notations of observed data
14 0.107 0.079 0.107 0.061 0.021 0.037 0.034 (yj, xj), yj = (yj, dj) xj = (x1j, x2j, . . ., xnj) (j = 1, 2, . . ., m) and
15 0.076 0.046 0.076 0.061 0.060 0.101 0.034 fuzzy coefficients (a, c), the regression model can be
16 0.120 0.079 0.120 0.061 0.021 0.022 0.034
mathematically written in the following LP problem:
17 0.106 0.079 0.116 0.103 0.035 0.022 0.034
18 0.019 0.027 0.019 0.032 0.102 0.022 0.034 X
m
min cjx j j
a;c
j¼1
The possibilistic regression equation is written as in the subject to
following: (4)
y j þ L1 ðaÞd j  ax j þ L1 ðaÞcjx j j
Y j ¼ A1 x1 j þ A2 x2 j þ . . . þ AK xk j ¼ Ax j y j  L1 ðaÞd j  ax j  L1 ðaÞcjx j j
(1) ð j ¼ 1; 2;    ; mÞ; c  0
x1 j ¼ 1; j ¼ 1; 2; . . . m
where L(u) is a membership function of a possibilistic regres-
where regression coefficient Ai is a triangular-shaped fuzzy
sion model and a is a possibility to which extent a sample is
number Ai = (ai, ci) with center ai and width ci. In the equation
included in the model.
(1), xj is the jth data.
Solving the LP problem mentioned above, we have the
According to the extension principal, we can rewrite Eq. (1)
possibilistic regression shown in Fig. 1. The relation (3)
Y j ¼ Ax j ¼ ðax j ; cjx j jÞ (2) between the model and the data is illustrated in the figure. In
other words, this possibilistic regression contains all data in its
where jxjj = (jx1jj,jx2jj, . . ., jxnjj). The output of the possibilistic width and results in expressing all possibilities which samples
regression Eq. (1), whose coefficients are fuzzy numbers, embody, that is, which the system treated should have. It is
results in a fuzzy number. possible in the formulation of the possibilistic regression model
Table 11
Comparison of R&D projects among three methods
AHP Conventional method Principal component analysis
Weight Ranking Score Ranking 1st PC 2nd PC Ranking
1 0.102 1 27 1 3.177 1.513 1
2 0.061 9 21 6 0.472 0.552 9
3 0.037 11 15 11 1.028 0.116 11
4 0.032 14 12 13 1.363 0.994 13
5 0.079 3 24 3 2.035 1.154 2
6 0.083 2 25 2 2.016 0.740 3
7 0.028 17 10 17 1.403 0.021 14
8 0.031 15 11 15 2.180 1.369 17
9 0.023 18 8 18 2.239 0.091 18
10 0.035 12 13 12 1.058 1.241 12
11 0.050 10 17 10 0.415 0.024 10
12 0.031 16 11 15 1.713 2.090 16
13 0.065 5 21 6 1.303 0.196 4
14 0.064 8 21 6 1.007 1.895 5
15 0.065 6 22 4 0.789 0.112 6
16 0.065 6 20 9 0.701 2.143 8
17 0.066 4 22 4 0.719 2.588 7
18 0.033 13 12 13 1.649 1.232 15
1272 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273

to treat non-fuzzy data with no width by setting the width dj to 0

in the above equations.

5.2. R&D evaluation model

As mentioned above, the total evaluation of an R&D

project is obtained in terms of Indices ‘‘Objective of R&D,’’
‘‘Necessity of R&D,’’ ‘‘Technological difficulty,’’ ‘‘Expected
sales volume’’ and ‘‘Possibility to get a patent.’’ On the
other hand, the cost of R&D is expressed using Index ‘‘Cost
of R&D.’’ But, ranking of an R&D project is mainly
decided by the total evaluation of R&D. By means of the
AHP, it is understood that Indices ‘‘Objective of R&D,’’ Fig. 3. Evaluation of R&D projects by R&D evaluation model.
‘‘Necessity of R&D,’’ ‘‘Expected sales volume’’ and
‘‘Possibility to get a patent’’ are most seriously influential
and next, Indices ‘‘Technological difficulty’’ and ‘‘Cost’’ are 6. Concluding remarks
weighed.
Here, we build an R&D evaluation model using a fuzzy The present days are named ‘‘age of knowledge’’ and a
regression model [6–11] on the basis of the above results. We company is evaluated with its intellectual property and effort to
employ the evaluated score of AHP as the total evaluation of develop a new market. The intellectual property, the company
each project (Table 11) and the evaluation indices as a has, is coming from its R&D activity. Therefore, in this paper,
explanatory value (Table 4). As a result, the following R&D we focused on a certain company and built the model to
evaluating model is obtained: evaluate an R&D project. We understood its latent structure
and the characteristics by means of a principal component
Y ¼ A 0 þ A1 X 1 þ A2 X 2 þ A3 X 3 þ A4 X 4 þ A 5 X 5 þ A6 X 6 analysis and the evaluation structure using AHP. Using these
results, R&D evaluation model which presume an evaluation
þ A7 X 7
score was built by a fuzzy regression model. Using the R&D
where a coefficient Ai (i = 0, 1, . . ., 7) has ai and ci as as a center evaluation model obtained by a fuzzy regression analysis, it
and a width of a triangular fuzzy number Ai = (ai, ci), respec- can describe the evaluation structure of the principal
tively. In addition, Since Indices 1 and 3 have the same value for component analysis and the AHP. It should be noted that on
all projects, it is not necessary to employ both indices in the basis of the R&D evaluation model, the structure of the
building the model. evaluation model of an R&D project is defined mainly using
The obtained fuzzy regression coefficients are shown in ‘‘Objective of R&D,’’ ‘‘Necessity of R&D’’ and ‘‘Possibility to
Table 12. The vagueness of Index 6 is rather larger than other get a patent.’’
indices because the coefficient c6, that is, a width of Index 6 is
large. There is difference between AHP and this model, but it References
should be noted that Indices ‘‘an objective of R&D,’’ ‘‘a
[1] S. Imoto, Y. Yabuuchi, J. Watada, Building a model to evaluate R&D
necessity of R&D’’ and ‘‘a possibility to get a patent’’ are projects, in: PICMT’05 Portland International Conference on Manage-
central in the evaluation. ment of Engineering and Technology, Portland, USA, July 31–August 4,
Fig. 3 shows an observed value and its predicted value. The 2005, (2005), pp. IM001–IM008.
predicted value is between the upper and lower bounds, and [2] S. Imoto, Y. Yabuuchi, J. Watada, The model’s of selection of project of
also the observed value is in the interval. In addition, evaluation research and development construction, in: 14th Intelligent System Sym-
posium, FAN Symposium’04 in Kochi, October 9, 2004, (2004), pp.
score fluctuates within the interval, it is found that these scores P237–P242.
should have a possibility of the change of their priority order. [3] C. Hayashi, T. Suzuki, Social Investigation and Quantity Conversion,
The relationship among them is clear from Fig. 3 concerning Iwanami Shoten, 1997 (in Japanese).
the R&D project. [4] C. Hayashi, T. Komazawa, Quantity Conversion Theory and Data Proces-
sing, Asakura Shoten, 1982 (in Japanese).
Table 12 [5] E. Kinoshita, et al., Theory and Actual of AHP, JUSE Press, Ltd., 2000(in
Fuzzy regression coefficients of R&D evaluation model Japanese).
[6] H. Tanaka, J. Watada, Possibilistic linear systems and their application
i ai ci to the linear regression model, Fuzzy Sets Syst. 27 (1988)
0 0.0040 0 275–289.
1 0.0048 0 [7] J. Watada, Y. Yabuuchi, Fuzzy robust regression analysis, in: Proceedings
2 0.0076 0 of FUZZ/IEEE’94: Third IEEE international Conference on Fuzzy Sys-
4 0.0001 0 tems, Florida, USA, June 26–July 2, (1994), pp. 1370–1376.
5 0.0024 0.0003 [8] J. Watada, Convex hull approach to efficient computation of fuzzy
6 0.0027 0.0010 regression analysis, in: Proceedings of The 4th Asia Fuzzy
7 0.0045 0 System Symposium, vol. 2, Tsukuba, June 2–3, (2000), pp. 1009–
1014.
 S. Imoto et al. / Applied Soft Computing 8 (2008) 1266–1273 1273

[9] J. Watada, Recent development of fuzzy regression model, in: Proceedings [11] Y. Yabuuchi, J. Watada, Possibilistic forecasting model and its application
of the 6th Czech-Japan Seminar, Valtice, Czech Republic, September 20– to analyze the economy in Japan, in: Proceedings of Eighth International
23, 2003. Conference on Knowledge-Based Intelligent Information & Engineering
[10] Y. Yabuuchi, J. Watada, Fuzzy robust regression analysis based on a Systems, Wellington, New Zealand, September 20–24, (2004), pp. III-
hyperelliptic function, J. Operations Res. Soc. Jpn. 39 (4) (1996) 512–524 151–III-158.
(in Japanese).