Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com
Procedia
Engineering
ProcediaProcedia
Engineering 00 (2011)
Engineering 000–000
29 (2012) 229 – 234
www.elsevier.com/locate/procedia

2012 International Workshop on Information and Electronics Engineering (IWIEE)

The Application of Generating Fuzzy ID3 Algorithm in

Performance Evaluation
Yan Lia*, Dandan Jiangb, Fachao Lia, b
a
College of Sciences, Hebei University of Science and Technology, Shijiazhuang 050018, China
b
School of Economics and Management, Hebei University of Science and Technology,
050018, Shijiazhuang, China

Abstract

Performance evaluation is an important part of human resource management and the core of performance
management, so the choice of evaluation methods is essential. The common methods are graphic rating scale, force
distribution, behaviorally anchored rating scale, management by objectives, 360-degree evaluation method and so on.
But these methods can not better handle the ambiguity and uncertainties of the data. To avoiding this problem, this
paper applies fuzzy ID3 algorithm which integrates the different fuzzy degree to employee performance evaluation,
so that to achieve a more effective and distinct classification in employee performance evaluation and provide a
reasonable foundation for implementing effective classification.

© 2011 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.

Keywords: Data mining, Generating fuzzy ID3 algorithm, Knowledge, Performance evaluation.

1. Introduction

Performance evaluation [1], is in according to predetermined criteria and certain evaluation procedures,
using scientific evaluation methods, to regular and irregular assessment and evaluation in working ability
and working performance of evaluation object. Effective and reasonable performance evaluation not only
mobilize the enthusiasm of the staffs and encourage them to continuously improve work efficiency, but
also find the problems of organizational personnel structure and promote rational and efficient allocation
of talent. This is an important basis for enterprises to establish reasonable and effective human resources
development strategy.

*
Yan Li. Tel.:13303112656.
E-mail address: byxliyan@163.com.

1877-7058 © 2011 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
doi:10.1016/j.proeng.2011.12.699
230
2 Yan
Y. Li et Li
al. et al. / Procedia
/ Procedia Engineering
Engineering 29 (2012)
00 (2011) 229 – 234
000–000

There are many methods of implementing performance evaluation, but these methods [2-4] can only
get some surface information of data, can not better tap the potential correlation between the data. data
mining methods represented by decision tree algorithm which are used in performance evaluation [5-7]
effectively compensate for this lack. But with the depth of this algorithm being applied, the researchers
found classical decision tree generation algorithm [8] can tap the potential correlation between the data,
but can not better handle the ambiguity and uncertainties of the data. While because fuzzy decision tree
[9-10] effectively integrates the advantages of fuzzy theory and decision trees, not only have a strong
decision analysis ability, but also can better handle the ambiguity and uncertainties of the data, at the
same time as the extension of decision tree, so it is applied to the performance evaluation [11].
It is worth noting, the current fuzzy decision tree induction algorithms do not systematically consider
the impact of different fuzzy levels, do not simply make uncertainty treatment awareness into the
selection of extended properties. To avoiding the lack of current algorithms, document [12] proposed a
fuzzy ID3 algorithm based on generating Hartley measure, this algorithm generally considered the
importance level and size of the cut set, then established a generating Hartley measure model with
structural characteristics in fuzzy sets and used it in uncertainty measure of fuzzy partition, effectively
reflected the different ambiguity degree effects on the classification. This paper uses this algorithm in
employees performance evaluation, the generated knowledge rules can be as an important reference for
business managers to implement staffs appraisal and management.

2. Fuzzy ID3 Algorithm Based on Generating Hartley Measure

Fuzzy ID3 algorithm [13] is an improvement of ID3 algorithm. In the recursive and structure process
of decision tree, fuzzy ID3 algorithm adopted the divide and conquer strategy, i.e. using the fuzzy
classification entropy of the attribute-value as heuristic function of expanded attributes, selecting the
smallest classification fuzzy entropy of the attribute as expanded attributes. Here, we make the
appointment: E={1, 2, � , N} denotes the training set, A={A1, A2, � , Am} represents condition
attribute set, Range(Ai)={Ai1, Ai2, � , Aimi } represents the range of the attribute Ai, Range(C)={c1,
c2, � , cm} the range of the decision attribute, then: 1) In fuzzy environment, attribute-value can be
considered as fuzzy set to the training set E={1, 2, � , N} (i.e. N dimensional fuzzy vector on E); 2) The
nodes of the decision tree can be considered as fuzzy set on E. According to the above analysis and
appointment, the following is the implementation process of fuzzy ID3 algorithm (where, for fuzzy set B
on E, M ( B ) = ∑i =1 B (i ) represents the cardinality of B):
∗ N

′
Step 1 For the current node D: a) For the given leaf criterion δ (0 < δ < 1) , if there is a k ∈ {1, 2,�, n}
such that f k ′ ( D) ≥ δ , then D is a leaf node; b) If father node of D has used all the condition attributes in
∗ ∗
branching process, then D is a leaf node. Where, f k ( D) = M ( D ∩ ck ) M ( D) .
Step 2 If D is not leaf node, then calculate the fuzzy classification entropies FE ( D, Ai ) of the condition
attributes Ai which are not used by father node of D in branching process, and select the smallest
classification fuzzy entropy to the attribute as expanded attributes( i.e. according to this attribute to
branch).
mij
FE ( D, Ai ) = − ∑ kji=1 E ( D ∩ Aij )
mi , (1)
n mijk mijk
E ( D ∩ Aij ) = − ∑ log 2
k =1 mij mij
. (2)
 YanY.Li
Lietetal.
al.//Procedia
ProcediaEngineering
Engineering29
00(2012)
(2011)229 – 234
000–000 231
3

m = M ∗ ( D ∩ Aij ) mi = ∑ kji=1 mij mijk = M ∗ ( D ∩ Aij ∩ ck ) mij = ∑ nk =1 mijk

Where, ij , , , .
Step 3 Repeat step 1 and step 2, until did not branch.
Step 4 Conversion the generating tree to the corresponding rules.
Actually, compared with the fuzzy ID3 algorithm, fuzzy ID3 algorithm based on generating Hartley
measure adjust mij in (1) to (3), adjust E ( D I Aij ) in (2) to (4).
1
=mij M
= ( D ∩ Aij ) ∫0 W (λ ) | ( D ∩ Aij )λ | dλ
, (3)
mijk
n
E ( D ∩ Aij ) =
−∑ [ H ( D ∩ Aij ∩ ck ) − H ( D ∩ Aij )]
k =1 mi
. (4)
1
( D ∩ Aij ∩ ck ) ∫0 W (λ ) | ( D ∩ Aij ∩ ck )λ | dλ H ( B) = ∫0 W (λ ) log 2 | Bλ | dλ
1
= m M=
Where, ijk , .
Obviously, after the adjustment, fuzzy ID3 algorithm based on generating Hartley measure is still the
extension of classical ID3 algorithm. W(λ) depicts the impact degree of cut-standard λ to the research
question. In reality, according to the properties of the research questions and different emphases, we can
select the specific form. At the same time, we can adjust the weight function W(λ) to reflect the
importance of different membership degree, then make uncertainty treatment awareness into the selection
of extended properties.
In order to facilitate description, fuzzy ID3 algorithm based on generating Hartley measure is called
generating fuzzy ID3 algorithm for short.

3. Performance Evaluation Case and Analysis

In a company, human resources department wants to imply assessment to the staffs of product sales
department. Condition attributes indexes are Morality, Ability, Attitude and Knowledge. Decision
attribute index is Performance. Now, it wants to assess a random sample of 20 staffs from 33 sales staffs,
the score values in the following Table 1.
For the above raw data, Now more commonly used approach was categorization to implement classify,
namely using the below percentile score standard to category: While score is more than or equal 85,
categorization is A (namely Excellent); While score is more than or equal 75 and less than 85,
categorization is B (namely Good); While score is more than or equal 65 and less than 75, categorization
is C (namely Ordinary); While score is less than 65, categorization is D (namely Poor). Categorization
conversion table is following as Table 2.
Obviously, in the process of the above index value categorization conversion, when a index value is 84,
it is classified as categorization C, while 85 which is only a difference of 1 with 84 is classified as
categorization B. Obviously it is unfair. Actually, for this kind of threshold, we can not clearly classify, so
it has a certain ambiguity. In practice, for ambiguity problem, using a clear decision tree is not up to
address of solving. So for the sort of performance evaluation problem, using vague language to describe
the process is more rational.

4. Performance Evaluation Model Based on Generating Fuzzy ID3 Algorithm

4.1. The fuzzy processing of data

In order to find the regularity between the data, we need to blur the original data. Now there are many
methods to do it, then considering the value characteristics of this instance, we use trapezoidal
232
4 Yan
Y. Li et Li
al. et al. / Procedia
/ Procedia Engineering
Engineering 29 (2012)
00 (2011) 229 – 234
000–000

membership function to blur the original data and calculate the corresponding membership, to unity the
data from quantitative to qualitative. Trapezoidal membership function is the following:
⎧ 0, 79 ≤ x ≤ 100 ⎧ 0, 89 ≤ x ≤ 100
⎧ 0, 69 ≤ x ≤100 ⎪ x − 71
71 ≤ x < 79 ⎪ x −81 ⎧ 1, 89 ≤ x ≤ 100
⎪ ⎪1 − 8 , ⎪1 − 8 , 81 ≤ x < 89
A(x) = ⎨1− x−861 , 61≤ x <= 69 , B( x) ⎪
69 ≤ x < 71
, ⎪ , D( x) ⎪ x −81 ,
= ⎨ 8 81 ≤ x < 89
.
⎨ 1, = C( x) ⎨ 1, 79 ≤ x < 81
⎪ 1, 0 ≤ x < 61 ⎪ x − 61 , ⎪ x −71 , ⎪ 0, 0 ≤ x < 81
⎩ 61 ≤ x < 69 71 ≤ x < 79 ⎩
⎪ 8 ⎪ 8
⎪⎩ 0, 0 ≤ x < 61 ⎪⎩ 0, 0 ≤ x < 71

（Poor） （Ordinary） （Good） （Excellent）

Table 1 The raw data of the sample set Table 2 The conversion data of the sample set
No. Morality Ability Attitude Knowledge Performance No. Morality Ability Attitude Knowledge Performance
1 71 81 74 68 73 1 C B C C C
2 73 58 61 94 71 2 C D D A C
3 82 63 59 94 72 3 B D D A C
4 71 83 89 84 82 4 C B A B B
5 70 73 73 87 74 5 C C C A C
6 69 92 65 92 79 6 C A C A B
7 91 70 74 90 81 7 A C C A B
8 56 93 88 78 79 8 D A A B B
9 78 73 61 75 72 9 B C D B C
10 90 76 90 90 88 10 A B A A A
11 79 92 79 92 85 11 B A B A B
12 69 58 83 91 73 12 C D B A C
13 73 64 70 83 70 13 C D C B C
14 83 87 63 82 79 14 B A D B B
15 80 63 69 90 73 15 B D C A C
16 57 65 75 61 64 16 D C B D C
17 84 80 70 93 82 17 B B C A B
18 82 55 55 73 66 18 B D D C C
19 57 66 74 93 72 19 D C C A C
20 55 72 81 73 70 20 D C B C C
According to the above trapezoidal membership function, we use fuzzy logic toolbox in software
Matlab to blur the original data, then get into the partial results is the following in Table 3.

4.2. The results of data mining and analysis

First, according to the classic Fuzzy ID3 algorithm, we do classification and generate decision tree for
the data in table 3. Corresponding decision tree is the following. When we use generating fuzzy ID3
algorithm to classify and generate decision tree for the data in table 3, corresponding decision tree is the
following(i.e. the level importance weight function ( W (λ ) = 2λ ). The related information (including the
number of node, the number of leaf, depth, average accuracy and assessment accuracy) from the above
decision tree is listed in the Table 4. We can know from the table, the value of the decision tree based on
generating fuzzy ID3 algorithm(i.e. the level importance weight function W (λ ) = 2λ ) is better than the
value of the decision tree based on fuzzy ID3 algorithm [14]. This shows, when selecting appropriate
weight function W(λ), that can generate the decision tree with more effective classification and higher
accuracy. Thus indicates that the generating fuzzy ID3 algorithm can be as an effective method in
implementing of employee performance evaluation. In addition, by these knowledge rule sets, enterprise
 YanY.Li
Lietetal.
al.//Procedia
ProcediaEngineering
Engineering29
00(2012)
(2011)229 – 234
000–000 233
5

management personnels can generally find out the basic situation and discipline of employees. At the
same time it also provides a reasonable foundation for implementing effective classification.

Figure 1 The decision tree based on classic fuzzy ID3 algorithm

Figure 2 The decision tree based on generating fuzzy ID3 algorithm

Table 3 The data after fuzzy processing
Morality Ability Attitude
No.
A B C D A B C D A B C D
1 0 0 1 0 0 1 0 0 0 0.375 0.625 0
2 0 0.25 0.75 0 0 0 0 1 0 0 0 1
3 0.125 0.875 0 0 0 0 0.25 0.75 0 0 0 1
4 0 0 1 0 0.25 0.75 0 0 1 0 0 0
5 0 0 1 0 0 0.25 0.75 0 0 0.25 0.75 0
6 0 0 1 0 1 0 0 0 0 0 0.5 0.5
7 1 0 0 0 0 0 1 0 0 0.375 0.625 0
8 0 0 0 1 1 0 0 0 0.875 0.125 0 0
9 0 0.875 0.125 0 0 0.25 0.75 0 0 0 0 1
10 1 0 0 0 0 0.625 0.375 0 1 0 0 0
11 0 1 0 0 1 0 0 0 0 1 0 0
12 0 0 1 0 0 0 0 1 0.25 0.75 0 0
13 0 0.25 0.75 0 0 0 0.375 0.625 0 0 1 0
14 0.25 0.75 0 0 0.75 0.25 0 0 0 0 0.25 0.75
15 0 1 0 0 0 0 0.25 0.75 0 0 1 0
16 0 0 0 1 0 0 0.5 0.5 0 0.5 0.5 0
17 0.375 0.625 0 0 0 1 0 0 0 0 1 0
18 0.125 0.875 0 0 0 0 0 1 0 0 0 1
19 0 0 0 1 0 0 0.625 0.375 0 0.375 0.625 0
20 0 0 0 1 0 0.125 0.875 0 0 1 0 0
Table 4 Data summary
Data The number of node The number of leaf Depth Average accuracy Assessment accuracy
Figure 2 9 28 4 87.40% 96%
Figure 3 5 15 4 94.10% 100%
234
6 Yan
Y. Li et Li
al. et al. / Procedia
/ Procedia Engineering
Engineering 29 (2012)
00 (2011) 229 – 234
000–000

5. Conclusions

Effective and reasonable performance evaluation not only mobilize the enthusiasm of the staff and
encourage them to continuously improve work efficiency, but also find the problems of organizational
personnel structure and promote rational and efficient allocation of talent. it is an important basis for
enterprises to establish reasonable and effective human resources development strategy. There were many
methods of implementing performance evaluation in the past, but these methods can not better handle the
ambiguity and uncertainties of the data, this paper applies fuzzy ID3 algorithm which integrates the
different fuzzy degree to employee performance evaluation. By defining the fuzzy membership functions
and blurring the original data, we generate fuzzy decision tree and the corresponding rule sets, so that to
achieve a more effective and distinct classification in employee performance evaluation and provide a
reasonable foundation for implementing effective classification. At the same time, this algorithm is
successfully applied, that explains the generating fuzzy ID3 algorithm may be used in a wider area of
space to some extent, e.g. customer satisfaction evaluation, customer relationship management, economic
analysis, rule extraction and so on. We will continue studies for these fields in the next step.

Acknowledgements

This work is supported by the National Natural Science Foundation of China (71071049, 70871036)
and the Natural Science Foundation of Hebei Province (F2011208056).

References

[1] Yonghong Cai. The status quo and reflection of performance evaluation. Beijing: Beijing Normal University Press; 2001, 4.
[2] Gary Dessler. Human resource management. 8th ed. New Jersey: Pretice-Hall Inc; 2000.
[3] T.A. Beerh Etal. Evaluation of 360 degree feedback ratings: relationship with each other and with performance and selection
predictors. Journal of Organization Behavior J. Organiz. Behav. 2001; 22: 775-788.
[4] Jeffrey·Mello. Strategic human resource management. Cincinnati: South-Western College Publishing; 2004, p. 246.
[5] Jinfeng He. Model and method of quality performance evaluation. Xian: Northwestern Polytechnical University; 2006.
[6]Yaomei Fang, Wangfeng He. The application of extension data mining in the university's teaching quality evaluation.
Practice and Understanding of Mathematics. 2009; 39(4): 82-87.
[7]Qin Qin, Qiuhong Duan, Yongqiang He. The data mining in researching performance evaluation. Journal of Hennan Institute
of Engineering (Natural Science Edition). 2010; 22(2): 60-62.
[8] J. R. Quinlan. Induction of decision trees. Machine Learning. 1986; 1(1): 81-106.
[9] Yufei Yuan, Michael J.Shaw. Induction of fuzzy decision trees. Fuzzy Sets and System. 1995; 69(1):125-139.
[10] X. G. Wang, D. S. Yeung and E. C. C. Tsang. A comparative study on heuristic algorithms for generating fuzzy decision
trees, IEEE Transactions on systems. Man and Cybernet. Ics-part B: Cybernetics. 2001; 31(2):215-226.
[11] Liming Zhao, Huaizhou Quan. The application of fuzzy number theory Ⅲ in human resource performance evaluation. The
Journal of Northwestern A&F University. 2004; 1(1):76.
[12] Fachao Li, Dandan Jiang. Fuzzy ID3 algorithm based on generating Hartley measure. Lecture Notes in Computer Science,
2011;6988: 188-196.
[13] M. Umanol, H. Okamoto, I. Hatono, H. Tamura, F. Kawachi, S. Umedzu, and J. Kinoshita. Fuzzy decision trees by fuzzy
ID3 algorithm and its application to diagnosis systems, in IEEE Int. Conf. on Fuzzy Systems. June 26–29, 1994; 2113–2118.
[14] Qin Liao, Chifeng Hao, Zhihong Chen. Data mining and mathematical modeling. Beijing: National Defence Industry
Publishing; 2010, p. 185.