You are on page 1of 12

Expert Systems

with Applications
Expert Systems with Applications 34 (2008) 399–410
www.elsevier.com/locate/eswa

Evaluating students’ learning achievement using fuzzy


membership functions and fuzzy rules
Shih-Ming Bai, Shyi-Ming Chen *

Department of Computer Science and Information Engineering, National Taiwan University of Science and Technology,
43, Section 4, Keelung Road, Taipei 106, Taiwan, ROC

Abstract

In recent years, some methods have been presented for applying the fuzzy set theory in educational grading systems. In this paper, we
present a new method for dealing with students’ learning achievement evaluation using fuzzy membership functions and fuzzy rules. The
proposed method considers the difficulty, importance and complexity of questions for students’ answerscripts evaluation. It provides a
useful way to distinguish the ranking order of students with the same score.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Fuzzy grading systems; Fuzzy reasoning; Grade membership functions; Students’ evaluation

1. Introduction ing assessment. Echauz and Vachtsevanos (1995) presented


a fuzzy grading system. Law (1996) presented a method for
Zadeh (1965) proposed the theory of fuzzy sets. In using fuzzy numbers in education grading systems. Ma
recent years, some researchers focused on the application and Zhou (2000) presented a fuzzy set approach to the
of the fuzzy set theory in educational grading systems assessment of student-centered learning. Wang and Chen
(Bai & Chen, 2006a, 2006b; Biswas, 1995; Chang & Sun, (2006a) presented two methods for evaluating the answer-
1993; Cheng & Yang, 1998; Chen & Lee, 1999; Chiang & scripts of students using fuzzy sets. Wang and Chen
Lin, 1994; Echauz & Vachtsevanos, 1995; Law, 1996; Ma (2006b) presented two methods for evaluating students’
& Zhou, 2000; Wang & Chen, 2006a, 2006b; Weon & answerscripts using fuzzy numbers associated with degrees
Kim, 2001; Wilson, Karr, & Freeman, 1998; Wu, 2003). of confidence. Wilson et al. (1998) presented a flexible,
Bai and Chen (2006a) presented a method for automat- adaptive and automatic fuzzy-based grade assigning sys-
ically constructing grade membership functions for stu- tem. Wu (2003) presented a method for applying the fuzzy
dents’ evaluation for fuzzy grading systems. Biswas set theory and the item response theory to evaluate the
(1995) presented a method for students’ evaluation using learning performance of students.
fuzzy sets. Chang and Sun (1993) presented a method for Weon and Kim (2001) pointed out that the chief aim of
fuzzy assessment of learning performance of junior high educational grading systems should consider the ‘‘diffi-
school students. Cheng and Yang (1998) presented a culty’’, ‘‘importance’’ and ‘‘complexity’’ for students’
method for using fuzzy sets in education grading systems. answerscripts evaluation. However, because the ‘‘diffi-
Chen and Lee (1999) presented two methods for students’ culty’’ is a very subjective parameter and to adjust the
evaluation using fuzzy sets. Chiang and Lin (1994) pre- scores of students during the evaluation process is not
sented a method for applying the fuzzy set theory to teach- appropriate, if we use the method presented in Weon and
Kim (2001) to evaluate students’ learning performance,
*
Corresponding author. Tel.: +886 2 27376417; fax: +886 2 27301081. the ranking order of the students may be different from
E-mail address: smchen@mail.ntust.edu.tw (S.-M. Chen). the original ones. Thus, the method presented in Weon

0957-4174/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2006.09.010
400 S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410

8
and Kim (2001) may cause some arguments about fairness > 1; if v 6 a;
>
>  2
in students’ learning performance evaluation. >
>
< 1  2 va
ca
; if a < v 6 b;
In this paper, we present a new method to deal with the lT ij ¼ ð3Þ
 2
learning achievement evaluation using fuzzy membership >
> vc
>
> 2 ca ; if b 6 v < c;
functions and fuzzy rules. The proposed method considers >
:
the difficulty, importance and complexity of questions for 0; if v P c;
students’ answerscript evaluation. It provides a useful where v denotes the question solving time for Pij, a denotes
way to distinguish the ranking order of the students with the permitted lower limit solving time for Pij, c denotes the
the same score. permitted upper limit solving time for Pij, and b ¼ aþc .
2
The rest of this paper is organized as follows. In Section (ii) Response accuracy considering the importance:
2, we briefly review Weon and Kim’s method for learning ‘‘Importance’’ is a criterion for indicating how much a stu-
achievement evaluation from Weon and Kim (2001). In dent can understand the question domain. If a student
Section 3, we present a new method to deal with the learn- works out a question that has a relatively high importance,
ing achievement evaluation using fuzzy membership func- then the response accuracy weighted factor increases; if a
tions and fuzzy rules. In Section 4, we use an example to student works out a question that has a relatively low
illustrate the process of dealing with the learning achieve- importance, then the response accuracy weighted factor
ment evaluation using fuzzy membership functions and decreases; if a student solves a question whose importance
fuzzy rules. The conclusions are discussed in Section 5. is average, then the initial response accuracy weighted fac-
tor is maintained. The fuzzy dilation method is used to
2. A review of Weon-and-Kim’s method for educational increase the weight factor and the fuzzy concentration
grading systems method is used to decrease the weight factor. Assume that
the response accuracy is considered, then we can get the
Weon and Kim (2001) pointed out that the chief aim of formula to compute the response accuracy ICOR(P) for
educational grading systems should consider the ‘‘diffi- the question domain P when ‘‘importance’’ is considered,
culty’’, ‘‘importance’’ and ‘‘complexity’’ for students’ shown as follows:
answerscripts evaluation. They present a method for edu- ICORðP Þ ¼ ðP 1 ; CORðP 1 ÞÞ; ðP 2 ; CORðP 2 ÞÞ; . . . ; ðP n ; CORðP n ÞÞ
cational grading systems using fuzzy sets. In the following, ( )
[n X
m
we briefly review Weon-and-Kim’s method for students’ ¼ P i; k
ðlP ij  lT ij Þ ; ð4Þ
answerscripts evaluation as follows. i¼1 j¼1

where k is the weighted factor; if a question is important,


2.1. Computation of response accuracy then we let k = 0.5; if the importance of a question is med-
ium, then we let k = 1; if a question is not important, then
(i) Response accuracy with limited time only: Assume that we let k = 2.
there are a series of questions P1, P2, . . . , Pn in a question (iii) Response accuracy considering the complexity:
domain P, ‘‘Complexity’’ is a criterion indicating whether a question
P ¼ P 1; P 2; . . . ; P n; ð1Þ is grouped under a COMPLEX category. The more com-
plex the question is, the more time the students spend to
where Pi denotes the ith question in P and 1 6 i 6 n. As- answer it. Because the time permission is a very important
sume that each question Pi has several sub-questions Pi1, - element, complexity-related questions are solved with the
Pi2, . . . , Pim, where 1 6 i 6 n, then the response accuracy time limit adjusted properly. Weon and Kim (2001) divided
COR(Pi) of Pi is defined as follows: the complexity-related conditions into three categories (i.e.,
COMPLEX, MEDIUM and SIMPLE) to get a standard
CORðP i Þ ¼ CORðP i1 ; P i2 ; . . . ; P im Þ
( ) deviation r, i.e., the time difference that each student
[n X
m spends answering a given question. We can increase the
¼ P i; ðlP ij  lT ij Þ ; ð2Þ time limit as much as the standard deviation r if the com-
i¼1 j¼1
plexity of a question is COMPLEX; we can decrease the
where Pi denotes the ith question on the answerscript, l de- time limit as much as the standard deviation r if the com-
notes the membership function of a fuzzy set, lP ij denotes plexity of a question is SIMPLE; we can maintain the ini-
the membership grade of the response accuracy of the jth tial time limit if the complexity of a question is MEDIUM.
sub-question of Pi, lP ij ¼ 1 means that the jth sub-question In this way, we can get the formula to compute the
of the ith question is completely correct, lP ij ¼ 0 means response accuracy CCOR(P) of the question domain P
that the when ‘‘complexity’’ is considered, shown as follows:
P jth sub-question of the ith question is completely
false, denotes the algebraic sum, · denotes the algebraic CCORðP Þ ¼ ðP 1 ; CORðP 1 ÞÞ; ðP 2 ; CORðP 2 ÞÞ; . . . ; ðP n ; CORðP n ÞÞ
product, lT ij denotes the membership grade of time that is ( )
[n X
m
k
needed to solve the problem Pij, and lT ij is calculated by the ¼ P i; ðlP ij  lT ij Þ ; ð5Þ
inverse sigmoid function, shown as follows: i¼1 j¼1
S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410 401

where lT ij is defined as follows: Weon and Kim (2001) used the following membership
8 functions to evaluate the learning achievement through
>
> 1; if v 6 a;
>
>  2 the response accuracy and the normalized values:
>
>
< 1  2 cva 0 a ; if a < v 6 b0 ;
lT ij ¼   ð6Þ \Very Good" ¼ x2 ; if x ¼ 1;
>
> 0 2
0
>
> 2 cvc ; if b 6 v < c ; 0
\Good" ¼ x; if 0 < x < 1;
>
>
0 a

: 0
0; if v P c ; 2x; if 0 < x 6 0:5;
\Medium" ¼
where v denotes the question solving time for Pij, a denotes 2x þ 2; if 0:5 < x < 1;
the permitted lower limit time for solving Pij, c 0 = c + r, \Bad" ¼ x; if 0 < x < 1;
0
and b0 ¼ aþc 2
. \Very Bad" ¼ ðx þ 1Þ ;
2
if x ¼ 0;
(iv) Response accuracy considering the difficulty: ‘‘Diffi-
culty’’ is a criterion for indicating how hard the question
is. The response accuracy weighted factor increases if a stu- where these membership function curves are shown in
dent works out a question that has a relatively high diffi- Fig. 1.
culty (i.e., DIFFICULT); the response accuracy weighted Each question is normalized and the normalized
factor decreases if a student works out a question that response accuracy is linguistically evaluated by one of the
has a relatively low difficulty (i.e., EASY); the initial previously defined membership functions after the response
response accuracy weighted factor is maintained if a stu- accuracy is computed. The normalized response accuracy
dent solves a problem whose difficulty is average (i.e., may belong to more than two membership functions. The
MEDIUM). The fuzzy dilation method is used to increase membership function that has the maximum membership
the weighted factor and the fuzzy concentration method is value is regarded as the suitable fuzzy linguistic variable.
used to decrease the weighted factor. Assume that the The process is shown as follows:
response accuracy is considered, then we can get a formula [
n
to compute the response accuracy DCOR(P) for the ques- EVALðP Þ ¼ fP i ; NORMðCORðP i ÞÞg
tion domain P when ‘‘difficulty’’ is considered, shown as i¼1
n 
[  
k

follows: ¼ P i ; FUZSETj MAX NORM ðCORðP i ÞÞ ;
j¼1
i¼1
DCORðP Þ ¼ ðP 1 ; CORðP 1 ÞÞ; ðP 2 ; CORðP 2 ÞÞ; . . . ; ðP n ; CORðP n ÞÞ
( ) ð16Þ
[n X
m
¼ P i; h
ðlP ij  lT ij Þ ; ð7Þ [
n
EVALðP Þ ¼ fP i ; NORMðICORðP i ÞÞg
i¼1 j¼1
i¼1
n 
[   
where h is the weighted factor; if the difficulty of the ques- ¼ P i ; FUZSETj MAX NORM ðICORðP i ÞÞ
k
;
tion is easy, then we let h = 1; if the difficulty of the ques- i¼1
j¼1

tion is medium, then we let h = 0.5; if the difficulty of the ð17Þ


question is high, then we let h = 0.25. [
n
EVALðP Þ ¼ fP i ; NORMðCCORðP i ÞÞg
i¼1
2.2. Evaluation functions and fuzzy evaluation n 
[  
k

¼ P i ; FUZSETj MAX NORM ðCCORðP i ÞÞ ;
j¼1
Because there are m sub-questions for each question, the i¼1

previously-gained response accuracy must fall in the inter- ð18Þ


val [0, 1]. Therefore, Weon and Kim (2001) do the normal-
ization process as follows:
Very Very
NORMðCORðP i ÞÞ ¼ CORðP i Þ=m; where 1 6 i 6 n; ð8Þ Bad Medium Good
1.0
NORMðICORðP i ÞÞ ¼ ICORðP i Þ=m; where 1 6 i 6 n; Bad Good
ð9Þ 0.8
NORMðCCORðP i ÞÞ ¼ CCORðP i Þ=m; where 1 6 i 6 n;
ð10Þ 0.6

NORMðDCORðP i ÞÞ ¼ DCORðP i Þ=m; where 1 6 i 6 n; 0.4


ð11Þ
NORMðCORðP ÞÞ ¼ CORðP Þ=ðm  nÞ; ð12Þ 0.2

NORMðICORðP ÞÞ ¼ ICORðP Þ=ðm  nÞ; ð13Þ


X
NORMðCCORðP ÞÞ ¼ CCORðP Þ=ðm  nÞ; ð14Þ 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

NORMðDCORðP ÞÞ ¼ DCORðP Þ=ðm  nÞ: ð15Þ Fig. 1. Fuzzy membership functions.


402 S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410

[
n
rate of the jth student Sj on the ith question Qi, tij 2 [0, 1],
EVALðP Þ ¼ fP i ; NORMðDCORðP i ÞÞg
i¼1 1 6 i 6 m and 1 6 j 6 n. Let G be a grade matrix storing
[n   
k
 the score of each question of a student, shown as follows:
¼ P i ; FUZSETj MAX NORM ðDCORðP i ÞÞ ; 2 3
j¼1
i¼1 Q1 g1
ð19Þ Q2 6 7
6 g2 7
G¼ . 6 . 76 ;
where FUZSETj(MAX(Pi)) denotes the fuzzy set that has .. 4 .. 7
5
the maximum membership grade when the normalized
value NORM(DCOR(Pi)) of accuracy COR(Pi) of prob- Qm gm
lem Pi belongs to several fuzzy sets. For more details,
where gi denotes the grade of the ith question Qi,
please refer to Weon and Kim (2001).
gi 2 [1, 100] and 1 6 i 6 m. Let IM be an importance matrix
However, because the ‘‘difficulty’’ is a very subjective
and C be a complexity matrix given by the domain expert
parameter, to adjust the scores of students during the eval-
for each question, respectively, shown as follows:
uation process is not appropriate. If we use the method pre-
sented in Weon and Kim (2001) to evaluate students’
learning performance, the ranking order of the students
may be different from the original ones. Thus, the method
presented in Weon and Kim (2001) may cause some argu-
ments about the fairness in students’ learning performance
evaluation. Because Weon-and-Kim’s method (2001) has
the above drawback, in the next section, we will propose
a new method for evaluating students’ learning perfor-
mance using fuzzy membership functions and fuzzy rules.
The proposed method considers the difficulty, importance
and complexity of questions for students’ answerscripts
evaluation. It provides a useful way to distinguish the rank-
ing order of students with the same score.

3. A new method for evaluating students’ learning


performance using fuzzy membership functions and where ImS1, ImS2, ImS3, ImS4 and ImS5 denote the impor-
fuzzy rules tance levels, respectively, ImS1 = ‘‘low’’, ImS2 = ‘‘more or
less low’’, ImS3 = ‘‘medium’’, ImS4 = ‘‘more or less high’’
In this section, we present a new method for students’ and ImS5 = ‘‘high’’; imij denotes the degree of membership
learning performance evaluation using fuzzy membership of the degree of importance of the ith question Qi belong-
functions and fuzzy rules. Assume there are m questions ing to the importance level ImSj, imij 2 [0, 1], 1 6 i 6 m and
and there are n students to answer these questions. Let Qi 1 6 j 6 5; CS1, CS2, CS3, CS4 and CS5 denote the com-
be the ith question on the answerscript and Sj be the jth plexity levels, respectively, CS1 = ‘‘low’’, CS2 = ‘‘more or
student, where 1 6 i 6 m and 1 6 j 6 n. Then, we can get less low’’, CS3 = ‘‘medium’’, CS4 = ‘‘more or less high’’
an accuracy rate matrix A and an answer-time-rate matrix and CS5 = ‘‘high’’; cij denotes the degree of membership
T, shown as follows: of the complexity of the ith question Qi belonging to the
complexity level CSj, cij 2 [0, 1], 1 6 i 6 m and 1 6 j 6 5.
Then, based on the accuracy rate matrix A and the grade
matrix G, we can calculate the total score TSj of each stu-
dent Sj, shown as follows:
X
m
TS j ¼ aij  gi ; ð20Þ
i¼1

where 1 6 j 6 n. If there are any students having the same


total grade, the proposed method can rank them properly.
The proposed method is now presented as follows:
Step 1: Based on the accuracy rate matrix A and the
answer-time-rate matrix T, calculate the average accuracy
rate AvgAi and the average answer-time-rate AvgTi for
each question Qi, shown as follows:
Pn
where aij denotes the accuracy rate of the jth student Sj on j¼1 aij
AvgAi ¼ ; ð21Þ
the ith question Qi, aij 2 [0, 1], tij denotes the answer-time- n
S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410 403
Pn
j¼1 t ij value of the grade of the ith question Qi belonging to the
AvgT i ¼ ; ð22Þ linguistic term FASj, faij 2 [0, 1], 1 6 i 6 m and 1 6 j 6 5;
n
FTS1, FTS2, FTS3, FTS4 and FTS5 denote the fuzzy terms
where 1 6 i 6 m. Then, fuzzify them based on the following
‘‘short’’, ‘‘more or less short’’, ‘‘medium’’, ‘‘more or less
five fuzzy sets ‘‘low’’, ‘‘more or less low’’, ‘‘medium’’,
long’’ and ‘‘long’’, respectively, ftij denotes the membership
‘‘more or less high’’ and ‘‘high’’ as shown in Fig. 2 and cal-
value of the average answer-time of the ith question Qi
culate their membership grades belonging to each fuzzy set,
belonging to the linguistic term FTSj, ftij 2 [0, 1], 1 6 i 6 m
respectively.
and 1 6 j 6 5.
Then, we can get the fuzzy grade matrix FA for the aver-
Step 2: Based on the fuzzy grade matrices FA, FT and
age accuracy and the fuzzy grade matrix FT for the average
fuzzy rules, perform the fuzzy reasoning to evaluate the dif-
answer-time, shown as follows:
ficulty of each question. The fuzzy rules we used for fuzzy
reasoning are represented by a fuzzy matrix as shown in
Table 1. We can let the accuracy rate and the answer-
time-rate have different weights, respectively. In this paper,
we let the weights of the accuracy rate and the answer-time-
rate be 0.6 and 0.4, respectively. Based on the fuzzy grade
matrices FA and FT, the fuzzy rule matrix shown in Table
1, and the weights of the accuracy rate and the answer-
time-rate, we can perform fuzzy reasoning to infer the dif-
ficulty of the question Qi represented by a vector
" #
low more or less low medium more or less high high
DQi ¼ ;
d i1 d i2 d i3 d i4 d i5

where 1 6 i 6 m, described as follows:


where FAS1, FAS2, FAS3, FAS4 and FAS5 denote the fuzzy (i) Based on Table 1, we can find the fuzzy rules to infer
terms ‘‘low’’, ‘‘more or less low’’, ‘‘medium’’, ‘‘more or less that the difficulty of the question Qi is ‘‘low’’, shown as
high’’ and ‘‘high’’, respectively, faij denotes the membership follows:

IF the accuracy is more or less high and the answer-


More or More or time-rate is short THEN the difficulty is low,
medium less high high
low less low IF the accuracy is high and the answer-time-rate is short
1.0
THEN the difficulty is low,
IF the accuracy is high and the answer-time-rate is more
0.8 or less short THEN the difficulty is low.

0.6 Based on these fuzzy rules, the formula for calculating di1 is
shown as follows:
0.4
d i1 ¼ Max½ð0:6  fai4 þ 0:4  fti1 Þ; ð0:6  fai5
0.2 þ 0:4  fti1 Þ; ð0:6  fai5 þ 0:4  fti2 Þ; ð23Þ

X where di1 denotes the degree of membership of the degree


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
of difficulty of the question Qi belonging to the fuzzy set
Fig. 2. Membership functions of the fuzzy sets ‘‘low’’, ‘‘more or less low’’, ‘‘low’’, di1 2 [0, 1], 1 6 i 6 m, and m denotes the number
‘‘medium’’, ‘‘more or less high’’ and ‘‘high’’. of questions in a student’s answerscript.

Table 1
A fuzzy rule matrix to infer the difficulty
Answer-time Accuracy
Low More or less low Medium More or less high High
Short Medium More or less low More or less low Low Low
More or less short More or less high Medium More or less low More or less low Low
Middle More or less high More or less high Medium More or less low More or less low
More or less long High More or less high More or less high Medium More or less low
Long High High More or less high More or less high Medium
404 S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410

(ii) Based on Table 1, we can find the fuzzy rules to infer where di3 denotes the degree of membership of the degree
that the difficulty of the question Qi is ‘‘more or less low’’, of difficulty of the question Qi belonging to the fuzzy set
shown as follows: ‘‘medium’’, di3 2 [0, 1], 1 6 i 6 m, and m denotes the num-
ber of questions in a student’s answerscript.
IF the accuracy is more or less low and the answer-time- (iv) Based on Table 1, we can find the fuzzy rules to infer
rate is short THEN the difficulty is more or less low, that the difficulty of the question Qi is ‘‘more or less high’’,
IF the accuracy is medium and the answer-time-rate is shown as follows:
short THEN the difficulty is more or less low,
IF the accuracy is medium and the answer-time-rate is IF the accuracy is low and the answer-time-rate is more
more or less short THEN the difficulty is more or less or less short THEN the difficulty is more or less high,
low, IF the accuracy is low and the answer-time-rate is mid-
IF the accuracy is more or less high and the answer- dle THEN the difficulty is more or less high,
time-rate is more or less short THEN the difficulty is IF the accuracy is more or less low and the answer-time-
more or less low, rate is middle THEN the difficulty is more or less high,
IF the accuracy is more or less high and the answer- IF the accuracy is more or less low and the answer-time-
time-rate is middle THEN the difficulty is more or less rate is more or less high THEN the difficulty is more or
low, less high,
IF the accuracy is high and the answer-time-rate is mid- IF the accuracy is medium and the answer-time-rate is
dle THEN the difficulty is more or less low, more or less high THEN the difficulty is more or less
IF the accuracy is high and the answer-time-rate is more high,
or less high THEN the difficulty is more or less low. IF the accuracy is medium and the answer-time-rate is
high THEN the difficulty is more or less high,
Based on these fuzzy rules, the formula for calculating di2 is IF the accuracy is more or less high and the answer-
shown as follows: time-rate is high THEN the difficulty is more or less
high.
d i2 ¼ Max½ð0:6  fai2 þ 0:4  fti1 Þ; ð0:6  fai3 þ 0:4  fti1 Þ;
ð0:6  fai3 þ 0:4  fti2 Þ; ð0:6  fai4 þ 0:4  fti2 Þ; Based on these fuzzy rules, the formula for calculating di4 is
ð0:6  fai4 þ 0:4  fti3 Þ; ð0:6  fai5 þ 0:4  fti3 Þ; shown as follows:
ð0:6  fai5 þ 0:4  fti4 Þ; ð24Þ d i4 ¼ Max½ð0:6  fai1 þ 0:4  fti2 Þ; ð0:6  fai1 þ 0:4  fti3 Þ;
ð0:6  fai2 þ 0:4  fti3 Þ; ð0:6  fai2 þ 0:4  fti4 Þ;
where di2 denotes the degree of membership of the degree
of difficulty of the question Qi belonging to the fuzzy set ð0:6  fai3 þ 0:4  fti4 Þ; ð0:6  fai3 þ 0:4  fti5 Þ;
‘‘more or less low’’, di2 2 [0, 1], 1 6 i 6 m, and m denotes ð0:6  fai4 þ 0:4  fti5 Þ; ð26Þ
the number of questions in a student’s answerscript.
(iii) Based on Table 1, we can find the fuzzy rules to infer where di4 denotes the degree of membership of the degree
that the difficulty of the question Qi is ‘‘medium’’, shown as of difficulty of the question Qi belonging to the fuzzy set
follows: ‘‘more or less high’’, di4 2 [0, 1], 1 6 i 6 m, and m denotes
the number of questions in a student’s answerscript.
IF the accuracy is low and the answer-time-rate is short (v) Based on Table 1, we can find the fuzzy rules to infer
THEN the difficulty is medium, that the difficulty of the question Qi is ‘‘high’’, shown as
IF the accuracy is more or less low and the answer-time- follows:
rate is more or less short THEN the difficulty is medium,
IF the accuracy is medium and the answer-time-rate is IF the accuracy is low and the answer-time-rate is more
middle THEN the difficulty is medium, or less high THEN the difficulty is high,
IF the accuracy is more or less high and the answer- IF the accuracy is low and the answer-time-rate is high
time-rate is more or less high THEN the difficulty is THEN the difficulty is high,
medium, IF the accuracy is more or less low and the answer-time-
IF the accuracy is high and the answer-time-rate is high rate is high THEN the difficulty is high.
THEN the difficulty is medium.
Based on these fuzzy rules, the formula for calculating di5 is
Based on these fuzzy rules, the formula for calculating di3 is shown as follows:
shown as follows: d i5 ¼ Max½ð0:6  fai1 þ 0:4  fti4 Þ; ð0:6  fai1 þ 0:4  fti5 Þ;
d i3 ¼ Max½ð0:6  fai1 þ 0:4  fti1 Þ; ð0:6  fai2 þ 0:4  fti2 Þ; ð0:6  fai2 þ 0:4  fti5 Þ; ð27Þ
ð0:6  fai3 þ 0:4  fti3 Þ; ð0:6  fai4 þ 0:4  fti4 Þ;
where di5 denotes the degree of membership of the degree
ð0:6  fai5 þ 0:4  fti5 Þ; ð25Þ of difficulty of the question Qi belonging to the fuzzy set
S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410 405

‘‘high’’, di5 2 [0, 1], 1 6 i 6 m, and m denotes the number of where aci1 denotes the degree of membership of the degree
questions in a student’s answerscript. of answer-cost of the question Qi belonging to the fuzzy set
Therefore, we can get the difficulty matrix D, shown as ‘‘low’’, aci1 2 [0, 1], 1 6 i 6 m, and m denotes the number of
follows: questions in a student’s answerscript.
(ii) Based on Table 2, we can find the fuzzy rules to infer
that the answer-cost of the question Qi is ‘‘more or less
low’’, shown as follows:

IF the difficulty is low and the complexity is middle


THEN the answer-cost is more or less low,
IF the difficulty is low and the complexity is more or less
high THEN the answer-cost is more or less low,
where DS1 = ‘‘low’’, DS2 = ‘‘more or less low’’, IF the difficulty is more or less low and the complexity is
DS3 = ‘‘medium’’, DS4 = ‘‘more or less high’’ and more or less low THEN the answer-cost is more or less
DS5 = high. low,
Step 3: Based on the difficulty matrices and the complex- IF the difficulty is more or less low and the complexity is
ity matrices, perform the fuzzy reasoning to evaluate the middle THEN the answer-cost is more or less low,
answer-cost of each question. The fuzzy rules we used for IF the difficulty is medium and the complexity is low
fuzzy reasoning are represented by a fuzzy matrix as shown THEN the answer-cost is more or less low,
in Table 2. We can let the difficulty and the complexity IF the difficulty is medium and the complexity is more or
have different weights, respectively. In this paper, we let less low THEN the answer-cost is more or less low,
the weights of the accuracy rate and the answer-time-rate IF the difficulty is more or less high and the complexity
be 0.7 and 0.3, respectively. Based on the fuzzy grade is low THEN the answer-cost is more or less low.
matrices D and C, the fuzzy rule matrix shown in Table
2, and the weights of the difficulty and the complexity, Based on these fuzzy rules, the formula for calculating aci2
we can perform fuzzy reasoning to infer the answer-cost is shown as follows:
of the question Qi represented by a vector
 aci2 ¼ Max½ð0:7  d i1 þ 0:3  ci3 Þ; ð0:7  d i1 þ 0:3  ci4 Þ;
low more or less low medium more or less high high
COQi ¼ ;
aci1 aci2 aci3 aci4 aci5 ð0:7  d i2 þ 0:3  ci2 Þ; ð0:7  d i2 þ 0:3  ci3 Þ;
ð0:7  d i3 þ 0:3  ci1 Þ; ð0:7  d i3 þ 0:3  ci2 Þ;
where 1 6 i 6 m, described as follows:
(i) Based on Table 2, we can find the fuzzy rules to infer ð0:7  d i4 þ 0:3  ci1 Þ; ð29Þ
that the answer-cost of the question Qi is ‘‘low’’, shown as
follows: where aci2 denotes the degree of membership of the degree
of answer-cost of the question Qi belonging to the fuzzy set
IF the difficulty is low and the complexity is low THEN ‘‘more or less low’’, aci2 2 [0, 1], 1 6 i 6 m, and m denotes
the answer-cost is low, the number of questions in a student’s answerscript.
IF the difficulty is low and the complexity is more or less (iii) Based on Table 2, we can find the fuzzy rules to infer
low THEN the answer-cost is low, that the answer-cost of the question Qi is ‘‘medium’’,
IF the difficulty is more or less low and the complexity is shown as follows:
low THEN the answer-cost is low.
IF the difficulty is low and the complexity is high THEN
Based on these fuzzy rules, the formula for calculating aci1 the answer-cost is more or less medium,
is shown as follows: IF the difficulty is more or less low and the complexity is
more or less high THEN the answer-cost is medium,
aci1 ¼ Max½ð0:7  d i1 þ 0:3  ci1 Þ; ð0:7  d i1 þ 0:3  ci2 Þ; IF the difficulty is medium and the complexity is middle
ð0:7  d i2 þ 0:3  ci1 Þ; ð28Þ THEN the answer-cost is medium,

Table 2
A fuzzy rule matrix to infer the answer-cost
Complexity Difficulty
Low More or less low Medium More or less high High
Low Low Low More or less low More or less low Medium
More or less low Low More or less low More or less low Medium More or less high
Middle More or less low More or less low Medium More or less high More or less high
More or less high More or less low Medium More or less high More or less high High
High Medium More or less high More or less high High High
406 S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410

IF the difficulty is more or less high and the complexity Based on these fuzzy rules, the formula for calculating aci5
is more or less low THEN the answer-cost is medium, is shown as follows:
IF the difficulty is high and the complexity is low THEN
the answer-cost is medium. aci5 ¼ Max½ð0:7  d i4 þ 0:3  ci5 Þ; ð0:7  d i5 þ 0:3  ci4 Þ;
ð0:7  d i5 þ 0:3  ci5 Þ; ð32Þ
Based on these fuzzy rules, the formula for calculating aci3
is shown as follows: where aci5 denotes the degree of membership of the degree
of answer-cost of the question Qi belonging to the fuzzy set
aci3 ¼ Max½ð0:7  d i1 þ 0:3  ci5 Þ; ð0:7  d i2 þ 0:3  ci4 Þ;
‘‘high’’, aci5 2 [0, 1], 1 6 i 6 m, and m denotes the number
ð0:7  d i3 þ 0:3  ci3 Þ; ð0:7  d i4 þ 0:3  ci2 Þ; of questions in a student’s answerscript.
ð0:7  d i5 þ 0:3  ci1 Þ; ð30Þ Therefore, we can get the answer-cost matrix CO shown
as follows:
where aci3 denotes the degree of membership of the degree
of answer-cost of the question Qi belonging to the fuzzy set
‘‘medium’’, aci3 2 [0, 1], 1 6 i 6 m, and m denotes the num-
ber of questions in a student’s answerscript.
(iv) Based on Table 2, we can find the fuzzy rules to infer
that the answer-cost of the question Qi is ‘‘more or less
high’’, shown as follows:

IF the difficulty is more or less low and the complexity is


high THEN the answer-cost is more or less high, where CoS1 = ‘‘low’’, CoS2 = ‘‘more or less low’’,
IF the difficulty is medium and the complexity is more or CoS3 = ‘‘medium’’, CoS4 = ‘‘more or less high’’ and
less high THEN the answer-cost is more or less high, CoS5 = high.
IF the difficulty is medium and the complexity is high Step 4: Based on the answer-cost matrices and the
THEN the answer-cost is more or less high, importance matrices, perform the fuzzy reasoning to eval-
IF the difficulty is more or less high and the complexity uate the adjustment value of each question. The fuzzy rules
is middle THEN the answer-cost is more or less high, we used for fuzzy reasoning are represented by a fuzzy
IF the difficulty is more or less high and the complexity matrix as shown in Table 3. We can let the importance
is more or less high THEN the answer-cost is more or and the answer-cost have the same weights, respectively.
less high, In this paper, we let the weights of the importance and
IF the difficulty is high and the complexity is more or the answer-cost be 0.5 and 0.5, respectively. Based on the
less low THEN the answer-cost is more or less high, fuzzy grade matrices IM and CO, the fuzzy rule matrix
IF the difficulty is high and the complexity is middle shown in Table 3, and the weights of the importance and
THEN the answer-cost is more or less high. the answer-cost, we can perform fuzzy reasoning to infer
the adjustment value of the question Qi represented by a
Based on these fuzzy rules, the formula for calculating aci4 vector

is shown as follows: low more or less low medium more or less high high
VQi ¼ ;
aci4 ¼ Max½ð0:7  d i2 þ 0:3  ci5 Þ; ð0:6  fai3 þ 0:4  fti4 Þ; vi1 vi2 vi3 vi4 vi5
ð0:6  fai3 þ 0:4  fti5 Þ; ð0:6  fai4 þ 0:4  fti3 Þ;
where 1 6 i 6 m, described as follows:
ð0:6  fai4 þ 0:4  fti4 Þ; ð0:6  fai5 þ 0:4  fti2 Þ;
(i) Based on Table 3, we can find the fuzzy rules to infer
ð0:6  fai5 þ 0:4  fti3 Þ; ð31Þ that the adjustment value of the question Qi is ‘‘low’’,
where aci4 denotes the degree of membership of the degree shown as follows:
of answer-cost of the question Qi belonging to the fuzzy set
‘‘more or less high’’, aci4 2 [0, 1], 1 6 i 6 m, and m denotes IF the importance is low and the answer-cost is low
the number of questions in a student’s answerscript. THEN the adjustment value is low,
(v) Based on Table 2, we can find the fuzzy rules to infer IF the importance is low and the answer-cost is more or
that the answer-cost of the question Qi is ‘‘high’’, shown as less low THEN the adjustment value is low,
follows: IF the importance is more or less low and the answer-
cost is low THEN the adjustment value is low.
IF the difficulty is more or less high and the complexity
is high THEN the answer-cost is high, Based on these fuzzy rules, the formula for calculating vi1 is
IF the difficulty is high and the complexity is more or shown as follows:
less high THEN the answer-cost is high,
vi1 ¼ Max½ð0:5  imi1 þ 0:5  aci1 Þ; ð0:5  imi1 þ 0:5  aci2 Þ;
IF the difficulty is high and the complexity is high
THEN the answer-cost is high. ð0:5  imi2 þ 0:5  aci1 Þ; ð33Þ
S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410 407

Table 3
A fuzzy rule matrix to infer the adjustment value
Answer-cost Importance
Low More or less low Medium More or less high High
Low Low Low More or less low More or less low Medium
More or less low Low More or less low More or less low Medium More or less high
Middle More or less low More or less low Medium More or less high More or less high
More or less high More or less low Medium More or less high More or less high High
High Medium More or less high More or less high High High

where vi1 denotes the degree of membership of the degree of IF the importance is more or less low and the answer-
adjustment value of the question Qi belonging to the fuzzy cost is more or less high THEN the adjustment value
set ‘‘low’’, vi1 2 [0, 1], 1 6 i 6 m, and m denotes the number is medium,
of questions in a student’s answerscript. IF the importance is medium and the answer-cost is mid-
(ii) Based on Table 3, we can find the fuzzy rules to infer dle THEN the adjustment value is medium,
that the adjustment value of the question Qi is ‘‘more or IF the importance is more or less high and the answer-
less low’’, shown as follows: cost is more or less low THEN the adjustment value is
medium,
IF the importance is low and the answer-cost is middle IF the importance is high and the answer-cost is low
THEN the adjustment value is more or less low, THEN the adjustment value is medium.
IF the importance is low and the answer-cost is more or
less high THEN the adjustment value is more or less Based on these fuzzy rules, the formula for calculating vi3 is
low, shown as follows:
IF the importance is more or less low and the answer-
cost is more or less low THEN the adjustment value is vi3 ¼ Max½ð0:5  imi1 þ 0:5  aci5 Þ; ð0:5  imi2 þ 0:5  aci4 Þ;
more or less low, ð0:5  imi3 þ 0:5  aci3 Þ; ð0:5  imi4 þ 0:5  aci2 Þ;
IF the importance is more or less low and the answer- ð0:5  imi5 þ 0:5  aci1 Þ; ð35Þ
cost is middle THEN the adjustment value is more or
less low,
IF the importance is medium and the answer-cost is low where vi3 denotes the degree of membership of the degree of
THEN the adjustment value is more or less low, adjustment value of the question Qi belonging to the fuzzy
IF the importance is medium and the answer-cost is set ‘‘medium’’, vi3 2 [0, 1], 1 6 i 6 m, and m denotes the
more or less low THEN the adjustment value is more number of questions in a student’s answerscript.
or less low, (iv) Based on Table 3, we can find the fuzzy rules to infer
IF the importance is more or less high and the answer- that the adjustment value of the question Qi is ‘‘more or
cost is low THEN the adjustment value is more or less less high’’, shown as follows:
low.
IF the importance is more or less low and the answer-
Based on these fuzzy rules, the formula for calculating vi2 is cost is high THEN the adjustment value is more or less
shown as follows: high,
IF the importance is medium and the answer-cost is
vi2 ¼ Max½ð0:5  imi1 þ 0:5  aci3 Þ; ð0:5  imi1 þ 0:5  aci4 Þ; more or less high THEN the adjustment value is more
ð0:5  imi2 þ 0:5  aci2 Þ; ð0:5  imi2 þ 0:5  aci3 Þ; or less high,
ð0:5  imi3 þ 0:5  aci1 Þ; ð0:5  imi3 þ 0:5  aci2 Þ; IF the importance is medium and the answer-cost is high
THEN the adjustment value is more or less high,
ð0:5  imi4 þ 0:5  aci1 Þ; ð34Þ
IF the importance is more or less high and the answer-
where vi2 denotes the degree of membership of the degree of cost is middle THEN the adjustment value is more or
adjustment value of the question Qi belonging to the fuzzy less high,
set ‘‘more or less low’’, vi2 2 [0, 1], 1 6 i 6 m, and m denotes IF the importance is more or less high and the answer-
the number of questions in a student’s answerscript. cost is more or less high THEN the adjustment value
(iii) Based on Table 3, we can find the fuzzy rules to infer is more or less high,
that the adjustment value of the question Qi is ‘‘medium’’, IF the importance is high and the answer-cost is more or
shown as follows: less low THEN the adjustment value is more or less
high,
IF the importance is low and the answer-cost is high IF the importance is high and the answer-cost is middle
THEN the adjustment value is more or less medium, THEN the adjustment value is more or less high.
408 S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410

Based on these fuzzy rules, the formula for calculating vi4 is ‘‘more or less high’’ and ‘‘high’’, respectively, advi denotes
shown as follows: the final adjustment value for question Qi, and 1 6 i 6 m.
Step 5: Assume there are k students having the same
vi4 ¼ Max½ð0:5  imi2 þ 0:5  aci5 Þ; ð0:5  imi3 þ 0:5  aci4 Þ; total grade, we construct a new grade matrix EA for these
ð0:5  imi3 þ 0:5  aci5 Þ; ð0:5  imi4 þ 0:5  aci3 Þ; equal-grade students, shown as follows:
ð0:5  imi4 þ 0:5  aci4 Þ; ð0:5  imi5 þ 0:5  aci2 Þ;
ð0:5  imi5 þ 0:5  aci3 Þ; ð36Þ

where vi4 denotes the degree of membership of the degree of


adjustment value of the question Qi belonging to the fuzzy
set ‘‘more or less high’’, vi4 2 [0, 1], 1 6 i 6 m, and m de-
notes the number of questions in a student’s answerscript.
(v) Based on Table 3, we can find the fuzzy rules to infer where eaij denotes the accuracy rate of the jth student ESj
that the adjustment value of the question Qi is ‘‘high’’, with respect to the ith question Qi, eaij 2 [0, 1], 1 6 i 6 m,
shown as follows: and 1 6 j 6 n.
Step 6: Based on the adjustment value advi, where
IF the importance is more or less high and the answer- 1 6 i 6 m, calculate the sum of difference for the students
cost is high THEN the adjustment value is high, with the same total grade. The formula for calculating
IF the importance is high and the answer-cost is more or the sum of difference SODj of the jth student is shown as
less high THEN the adjustment value is high, follows:
IF the importance is high and the answer-cost is high
THEN the adjustment value is high.
k X
X m
SODj ¼ ðeaij  eaip Þ  gi  ð0:5 þ advi Þ; ð39Þ
p¼1 i¼1
Based on these fuzzy rules, the formula for calculating vi5 is
shown as follows: where 1 6 j 6 k. Then, sort the values of SODj in a
descending sequence, where 1 6 j 6 k, to get the new rank-
vi5 ¼ Max½ð0:5  imi4 þ 0:5  aci5 Þ; ð0:5  imi5 þ 0:5  aci4 Þ; ing order of the students with the same total grade.
ð0:5  imi5 þ 0:5  aci5 Þ; ð37Þ
4. An example
where vi5 denotes the degree of membership of the degree of
adjustment value of the question Qi belonging to the fuzzy In this section, we use an example to illustrate the pro-
set ‘‘high’’, vi5 2 [0, 1], 1 6 i 6 m, and m denotes the num- cess of the students’ learning performance evaluation using
ber of questions in a student’s answerscript. fuzzy membership functions and fuzzy rules. Assume that
Therefore, we can get the adjustment matrix V as there are five questions Q1, Q2, Q3, Q4, Q5 on the answer-
follows: script and assume that there are ten students S1, S2, . . . , S10
to answer these questions. Assume that the accuracy rate
matrix A, the answer-time-rate matrix T, the grade matrix
G, the importance matrix IM and the complexity matrix C
are as follows:

where VS1 = ‘‘low’’, VS2 = ‘‘more or less low’’,


VS3 = ‘‘medium’’, VS4 = ‘‘more or less high’’ and
VS5 = high. Based on the adjustment matrix V, we perform
the following operation to evaluate the final adjustment
value advi of the question Qi:

0:1  vi1 þ 0:3  vi2 þ 0:5  vi3 þ 0:7  vi4 þ 0:9  vi5
advi ¼ ;
0:1 þ 0:3 þ 0:5 þ 0:7 þ 0:9
ð38Þ

where 0.1, 0.3, 0.5, 0.7 and 0.9 are the defuzzification values
of the fuzzy sets ‘‘low’’, ‘‘more or less low’’, ‘‘medium’’,
S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410 409
2 3
Q1 10 grade matrix FT for the average answer-time, shown as
6 7 follows:
Q2 6 15 7
6 7
G ¼ Q3 6 7
6 20 7;
6 7
Q4 4 25 5
Q5 30

[Step 2] Based on the fuzzy grade matrices FA and FT, the


fuzzy rule matrix shown in Table 1 and formulas (23)–(27),
we can perform fuzzy reasoning to evaluate the difficulty of
each question. Then, we can get the difficulty matrix D,
Then, based on the accuracy rate matrix A, the grade ma- shown as follows:
trix G and formula (20), we can get the total grade of each
student, where the total grades of the students S1, S2, S3,
S4, S5, S6, S7, S8, S9 and S10 are 67.6, 54.05, 38.4, 49.7,
49.7, 48.8, 46.1, 52.3, 85.95 and 49.7, respectively. Because
the students S4, S5 and S10 have the same total grade, they
get the same ranking order (i.e., Top 5), i.e., the ranking
order of these ten students is follows:
S 9 > S 1 > S 2 > S 8 > S 4 ¼ S 5 ¼ S 10 > S 6 > S 7 > S 3 : [Step 3] Based on the difficulty matrix D and the complexity
matrix C, the fuzzy rule matrix shown in Table 2 and for-
Then, we use the proposed method to deal with the ranking
mulas (28)–(32), we can perform fuzzy reasoning to evalu-
order of the students S4, S5 and S10, shown as follows:
ate the answer-cost for each question. Then, we can get the
[Step 1] Based on the accuracy rate matrix A, the answer-
answer-cost matrix CO, shown as follows:
time-rate matrix T, formulas (21) and (22), we can calculate
the average accuracy rate AvgAi and the average answer-
time-rate AvgTi for each question Qi, 1 6 i 6 5, shown as
follows:
ð0:59 þ 0:35 þ 1 þ 0:66 þ 0:11 þ 0:08 þ 0:84 þ 0:23 þ 0:4 þ 0:24Þ
AvgA1 ¼ ¼ 0:45;
10
ð0:01 þ 0:27 þ 0:14 þ 0:04 þ 0:88 þ 0:16 þ 0:04 þ 0:22 þ 0:81 þ 0:53Þ
AvgA2 ¼ ¼ 0:31;
10
ð0:77 þ 0:69 þ 0:97 þ 0:71 þ 0:17 þ 0:86 þ 0:87 þ 0:42 þ 0:91 þ 0:74Þ
AvgA3 ¼ ¼ 0:711;
10
AvgA4 ¼
ð0:73 þ 0:72 þ 0:18 þ 0:16 þ 0:5 þ 0:02 þ 0:32 þ 0:92 þ 0:9 þ 0:25Þ
¼ 0:47; [Step 4] Based on the answer-cost matrix CO and the impor-
10
ð0:93 þ 0:49 þ 0:08 þ 0:81 þ 0:65 þ 0:93 þ 0:39 þ 0:51 þ 0:97 þ 0:61Þ tance matrix IM, the fuzzy rule matrix shown in Table 3 and
AvgA5 ¼ ¼ 0:637;
10 formulas (33)–(37), we can perform fuzzy reasoning to eval-
ð0:7 þ 0:4 þ 0:1 þ 1 þ 0:7 þ 0:2 þ 0:7 þ 0:6 þ 0:4 þ 0:9Þ
AvgT 1 ¼ ¼ 0:57; uate the adjustment value for each question. Then, we can
10
ð1 þ 0 þ 0:9 þ 0:3 þ 1 þ 0:3 þ 0:2 þ 0:8 þ 0 þ 0:3Þ get the adjustment matrix V, shown as follows:
AvgT 2 ¼ ¼ 0:48;
10
ð0 þ 0:1 þ 0 þ 0:1 þ 0:9 þ 1 þ 0:2 þ 0:3 þ 0:1 þ 0:4Þ
AvgT 3 ¼ ¼ 0:31;
10
ð0:2 þ 0:1 þ 0 þ 1 þ 1 þ 0:3 þ 0:4 þ 0:8 þ 0:7 þ 0:5Þ
AvgT 4 ¼ ¼ 0:5;
10
ð0 þ 0:1 þ 1 þ 1 þ 0:6 þ 1 þ 0:8 þ 0:2 þ 0:8 þ 0:2Þ
AvgT 5 ¼ ¼ 0:57:
10

Then, after fuzzifying these values, we can get the fuzzy


grade matrix FA for the average accuracy and the fuzzy
410 S.-M. Bai, S.-M. Chen / Expert Systems with Applications 34 (2008) 399–410

Then, based on the adjustment matrix V and formula (38), considers the difficulty, importance and complexity of
we can calculate the final adjustment value advi for each questions for students’ answerscript evaluation. It provides
question Qi, where 1 6 i 6 5, shown as follows: a useful way to distinguish the ranking order of students
ð0:1  0:38 þ 0:3  0:38 þ 0:5  0:66 þ 0:7  0:88 þ 0:9  0:75Þ with the same score.
adv1 ¼ ¼ 0:71;
0:1 þ 0:3 þ 0:5 þ 0:7 þ 0:9
ð0:1  0:36 þ 0:3  0:66 þ 0:5  0:66 þ 0:7  0:76 þ 0:9  0:43Þ Acknowledgements
adv2 ¼ ¼ 0:59;
0:1 þ 0:3 þ 0:5 þ 0:7 þ 0:9
ð0:1  0:33 þ 0:3  0:43 þ 0:5  0:76 þ 0:7  0:86 þ 0:9  0:80Þ
adv3 ¼
0:1 þ 0:3 þ 0:5 þ 0:7 þ 0:9
¼ 0:75; This work was supported in part by the National Sci-
ð0:1  0:90 þ 0:3  0:88 þ 0:5  0:68 þ 0:7  0:40 þ 0:9  0:32Þ ence Council, Republic of China, under Grant NSC 94-
adv4 ¼ ¼ 0:51;
0:1 þ 0:3 þ 0:5 þ 0:7 þ 0:9 2213-E-011-003.
ð0:1  0:34 þ 0:3  0:80 þ 0:5  0:76 þ 0:7  0:71 þ 0:9  0:25Þ
adv5 ¼ ¼ 0:55:
0:1 þ 0:3 þ 0:5 þ 0:7 þ 0:9
References
[Step 5] Because there are 3 students with the same total
Bai, S. M., & Chen, S. M. (2006a). Automatically constructing grade
grade, we construct a new grade matrix EA for the equal-
membership functions for students’ evaluation for fuzzy grading
grade students S4, S5 and S10, shown as follows: systems. In Proceedings of the 2006 world automation congress,
Budapest, Hungary.
Bai, S. M., & Chen, S. M. (2006b). A new method for students’ learning
achievement using fuzzy membership functions. In Proceedings of the
11th conference on artificial intelligence, Kaohsiung, Taiwan, Republic of
China.
Biswas, R. (1995). An application of fuzzy sets in students’ evaluation.
Fuzzy Sets and Systems, 74(2), 187–194.
Chang, D. F., & Sun, C. M. (1993). Fuzzy assessment of learning
performance of junior high school students. In Proceedings of the 1993
first national symposium on fuzzy theory and applications, Hsinchu,
where ES1 denotes the student S4, ES2 denotes the student Taiwan, Republic of China (pp. 1–10).
S5, and ES3 denotes the student S10. Cheng, C. H., & Yang, K. L. (1998). Using fuzzy sets in education grading
[Step 6] Based on the final adjustment values adv1, adv2, system. Journal of Chinese Fuzzy Systems Association, 4(2), 81–89.
adv3, adv4 and adv5, we can calculate the sum of difference Chen, S. M., & Lee, C. H. (1999). New methods for students’ evaluating
SOD1 for the student ES1, shown as follows: using fuzzy sets. Fuzzy Sets and Systems, 104(2), 209–218.
Chiang, T. T., & Lin, C. M. (1994). Application of fuzzy theory to
SOD1 ¼ ð0:66  0:11 þ 0:66  0:24Þ  10  ð0:5 þ 0:71Þ teaching assessment. In Proceedings of the 1994 second national
conference on fuzzy theory and applications, Taipei, Taiwan, Republic
þ ð0:04  0:88 þ 0:04  0:53Þ  15  ð0:5 þ 0:59Þ of China (pp. 92–97).
Echauz, J. R., & Vachtsevanos, G. J. (1995). Fuzzy grading system. IEEE
þ ð0:71  0:17 þ 0:71  0:74Þ  20  ð0:5 þ 0:75Þ Transactions on Education, 38(2), 158–165.
þ ð0:16  0:5 þ 0:16  0:25Þ  25  ð0:5 þ 0:51Þ Law, C. K. (1996). Using fuzzy numbers in education grading system.
Fuzzy Sets and Systems, 83(3), 311–323.
þ ð0:81  0:65 þ 0:81  0:61Þ  30  ð0:5 þ 0:55Þ Ma, J., & Zhou, D. (2000). Fuzzy set approach to the assessment of
student-centered learning. IEEE Transactions on Education, 43(2),
¼ 3:15: 237–241.
Wang, H. Y., & Chen, S. M. (2006a). New methods for evaluating the
In the same way, we can get the sum of difference answerscripts of students using fuzzy sets. In Proceedings of the 19th
SOD2 = 5.3 for the student S5 and the sum of difference international conference on industrial, engineering & other applications
of applied intelligent systems, Annecy, France (pp. 442–451).
SOD3 = 2.15 for the student S10. Because SOD1 >
Wang, H. Y., & Chen, S. M. (2006b). New methods for evaluating
SOD3 > SOD2, we can see that the ranking order of the students’ answerscripts using fuzzy numbers associated with degrees of
students S4, S5 and S10 is: S4 > S10 > S5. That is, the stu- confidence. In Proceedings of the 2006 IEEE international conference on
dents S4, S10 and S5 are the Top 5, the Top 6 and the fuzzy systems, Vancouver, BC, Canada (pp. 5492–5497).
Top 7, respectively. Therefore, the ranking order of these Weon, S., & Kim, J. (2001). Learning achievement evaluation strategy
using fuzzy membership function. In Proceedings of the 31st ASEE/
ten students becomes:
IEEE frontiers in education conference, Reno, NV (Vol. 1, pp. 19–24).
S 9 > S 1 > S 2 > S 8 > S 4 > S 10 > S 5 > S 6 > S 7 > S 3 : Wilson, E., Karr, C. L., & Freeman, L. M. (1998). Flexible, adaptive,
automatic fuzzy-based grade assigning system. In Proceedings of the
1998 north American fuzzy information processing society (NAFIPS)
5. Conclusions conference (pp. 334–338).
Wu, M. H. (2003). Research on applying fuzzy set theory and item
response theory to evaluate learning performance. Master Thesis,
In this paper, we have presented a new method to deal Department of Information Management, Chaoyang University of
with the learning achievement evaluation using fuzzy mem- Technology, Wufeng, Taichung County, Taiwan, Republic of China.
bership functions and fuzzy rules. The proposed method Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8, 338–353.

You might also like