Sei sulla pagina 1di 6

2018 International Conference on Applied Information Technology and Innovation (ICAITI)

Comparative Analysis of Multi-Criteria Decision


Making for Student Degree Completion Time based
on Entropy Weighted
Masna Wati Herman Santoso Pakpahan Niken Novirasari
dept. Computer Science and Information dept. Computer Science and Information dept. Computer Science and Information
Technology Technology Technology
Mulawarman University Mulawarman University Mulawarman University
Samarinda, Indonesia Samarinda, Indonesia Samarinda, Indonesia
masnawati.ssi@gmail.com pakpahan.herman891@gmail.com nikennovirasari@gmail.com

Abstract— Undergraduate student’s criteria such as grade based on the measurement of uncertain information
point average, high school major, age when they registered, contained in the decision matrix which directly generates a
etc., which exist in academic data can be utilized to predict the set of weights for a given criterion based on mutual contrast
student degree-completion time. This paper tries to get the of individual criteria values of variants for each criteria and
weight of ten criteria objectively using entropy and used those then for all the criteria at the same time [4]. Entropy is the
criteria weighting in comparative analysis for prediction of amount of information indicating the size of impurities. The
accuracy and precision of three multi-criteria decision-making impurity measure can determine how informative an attribute
methods, i.e., Simple Additive Weighting (SAW), Weighted input is in the form of sub criteria. Furthermore, this research
Product (WP), and Technique for Order of Preference by
uses the entropy value of each sub criteria as the weight of
Similarity to Ideal Solution (TOPSIS). The result showed that
the fourth grade point average has the biggest impact on
the sub criteria. While gain is the result of attribute selected
degree-completion time with 0.1589 of weight and weighted using entropy to determine the best attribute. The greater the
product is the best method to use with percentage of accuracy gain of a criteria the is, the more influential the criteria
61.3 percent and precision 52.85 percent. The result of this becomes [5].
research can be used to help academic staff in making decision Therefore this study will compare the results of each
for student admission and administering mentoring process. method in multi-criteria decision-making using the case
studies of graduate’s degree-completion time in Faculty of
Keywords— student, degree-completion, multi-criteria,
entropy, decision making, accuracy, precision
Computer Science and Information Technology (FKTI),
Mulawarman University (Unmul) as comparative dataset.
The purpose of this research is to find out the weight of each
I. INTRODUCTION criteria and the most eligible method of MCDM for decision
Decision-making in government, business, and education support system in FKTI Unmul student’s ideal degree-
has evolved from decisions taken subjectively by leaders into completion time.
decisions taken objectively based on data and from decisions
taken under one criteria (i.e., profit) to be based on many II. RELATED WORKS
criteria [1]. The amount of data required and influential
criteria encourage the use of computer systems to process Based on the prediction of degree-completion time of
data into a decision. This system is called decision support students obtained from alumni data of FKTI Unmul from
system. Decision support systems that use multiple criteria 2014 to 2016, the degree-completion time is influenced by
are called multi-criteria decision support systems (MCDM). eleven (11) criteria i.e. student’s gender, student’s age when
MCDM is used to establish the best alternative of a number they are registered in university, student’s place of birth,
of alternatives based on certain criteria [2]. student’s high school major, student’s high school basic
funding source, student’s 1st semester GPA, student’s 2nd
The current problem is that in one case study there are semester GPA, student’s 3rd semester GPA, student’s 4th
many MCDM methods that can be used. Therefore, previous semester GPA, student’s participation in non-academic
researchers conducted a comparison of methods before organization, and student’s degree-completion time [6].
implementing them on decision support systems. However,
determining the method to be used is based solely on the Implementation of simple additive weighting (SAW) and
similarity of the process or the popularity of the method weighted product (WP) on decision support system for new
rather than the alternative data form and the available criteria employee acceptance at PT Warta Medi Nusantara resulted
[3]. Despite the fact that comparative research results will be that WP method is more detailed than SAW method but
more ideal if all possible methods are used to be tested on SAW method has faster execution time than WP method [6].
alternative data forms and criteria. In comparison research, simple additive weighting and
Furthermore the criteria weight is determined TOPSIS for Open Recruitment of Informatics Laboratory
subjectively by decision maker, so sometimes the weight is Students at Trunojoyo Madura University resulted in a
not accurate. In this paper, the criteria’s weight will be sequence match with real result of 33.33% for SAW method
determined using entropy and gain values. The use of and 46.67% for TOPSIS [8].
entropy method in determining objective criteria weights is

56
Furthermore, in the selection of simplicial election using The TABLE I shows us weighted criteria and weighted
Weighted Product (WP) method and Simple Additive sub-criteria determined by the chance of gain for criteria and
Weighting (SAW) method yields accuracy for WP method entropy for sub-criteria for each student’s sub-criteria and
with 89% and accuracy test result for SAW method with criteria.
89%. While computational time test shows that SAW
method has faster execution time than WP method [9].
From the previous research, weighted product and simple TABLE I. CRTERIA’S WEIGHT AND SUB-CRITERIA’S WEIGHT
additive weighting have the same accuracy but WP is more
accurate than SAW and SAW is faster than WP. While Criteria Value Sum F M S Entropy G
Technique for Order of Preference by Similarity to Ideal
Sum 450 95 98 257 1.414
Solution (TOPSIS) has a higher accuracy than SAW but the
accuracy is still weak with less than 50%. Then, this study 0.023
will compare SAW, WP, and TOPSIS in the case study of Gender Male 314 51 70 193 1.34
degree-completion time in FKTI Unmul using weighted
criteria which are objectively generated from the chance of Female 136 44 28 64 1.508
entropy and gain. The results of this study will determine the 0.022
best method for building the decision support system for <18
degree-completion time of FKTI Unmul students. years 68 8 10 50 1.096
old
Age 18
III. MULTI-CRITERIA DECSION MAKING PROCESS
years 274 68 56 150 1.443
It is necessary for each decision-making system to old
determine the method to be used. The method chosen >18
depends on the form of criteria. The weight of objective years 108 19 32 57 1.448
old
criteria can be determined by the chance of entropy and gain.
0.002

A. Alternative Dataset Univer


sity 159 34 39 86 1.453
Dataset in this paper uses graduate’s data of degree- Place place
completion time in Faculty of Computer Science and birth Non-
Information Technology (FKTI), Mulawarman University univers
291 61 59 171 1.39
ity
(Unmul) from July 2015 to December 2017. There are 450
place
valid dataset with 95 of fast graduates, 98 of moderate
0.004
graduates, and 257 of slow graduates. High
school SMA 293 66 58 169 1.405
B. Objective Weight of Student’s Criteria and Sub-Criteria major
SMK 157 29 40 88 1.421
Entropy and gain are used to determine weight. Entropy
0.006
is the amount of information indicating the size of impurities. High
The impurity measure can determine how informative an school State 388 81 79 228 1.39
attribute input is in the form of sub criteria. Furthermore, this funding
Private 62 14 19 29 1.52
research uses the entropy value of each sub criteria as the
weight of the sub criteria. While gain is the result of attribute 0.035
selected using entropy to determine the best attribute. The 1.6-
17 2 2 13 1.022
greater the gain of a criteria is, the more influence this 2.59
1st GPA
2.6-
criteria has [5]. 3.59
318 58 61 199 1.328

The following equation (1) shows us how to get entropy 3.6-4 115 35 35 45 1.574
value and equation (2) shows us gain value:
0.136
0.0-
6 0 0 6 0
 ( ) ∑   1.59
1.6-
2nd GPA 31 2 0 29 0
2.59
In equation (1), S means the set of the cases, n is count of 2.6-
S partition, and pi is Si to S proportion. For example in 3.6-4 270 39 59 172 1.297
3.59
of 4th GPA there are 70, 71, 81 for Si (i= 1,2,3) and 222 for 3.6-4 143 54 39 50 1.572
S.
0.07
0.0-
1 0 0 1 0
 ( ) ( ) ∑ | | ( )  1.59
1.6-
3rd GPA 19 1 0 18 0
2.59
In equation (2), S means the set of the cases, A is 2.6-
attribute, n is count of A attribute partition, |Si| is count of 335 60 72 203 1.359
3.59
cases in the i partition, and |S| is count of S cases. For 3.6-4 95 34 26 35 1.573
example in the 4th GPA there are 1414 of entropy S reduced th
4 GPA 0.159
by summing 19, 17, 192, 222 of Si (i=1,2,3) by the 450 and
multiplying with entropy Si.

57
Criteria Value Sum F M S Entropy G degree. In figure 1, it is shown that criteria of the 4th GPA
0.0- and sub-criteria 3.6 to 4 have the most influence for degree-
19 0 0 19 0 completion time. It is important to note that further in multi-
1.59
1.6- criteria decision-making, not only sub-criteria but also
17 0 0 17 0
2.59 criteria itself influence degree-completion time.
2.6-
192 25 27 140 1.113
3.59
C. Simple Additive Weighting (SAW)
3.6-4 222 70 71 81 1.582
The basic logic of simple additive weighting is to obtain
0.008 a weighted sum of the performance ratings of each
Organiza Join 31 8 11 12 1.565 alternative over all attributes. The greater the value of the
tion preference is, the more preferred the alternative is [10]. The
Non-
419 87 87 245 1.394 dataset must be set in the decision matrix (X) based on the
join
match with entropy value, for example a student with 3.83 of
Below is detailed description from Table I: the 4th GPA, the matrix must be set to value 1.582. In
addition, SAW method normally requires the decision matrix
 F: fast degree-completion time in less than four years to be normalized in order to have a comparable scale for all
ratings in X by equation (3):
 M: moderate degree-completion time in four years
 S: slow degree-completion time in more than four
years
 {  
 G: gain value for each criteria
The weight of each sub-criteria produced with entropy
formula is shown on equation (1). Whereas the weight of The overall preference value of each alternative (Vi) is
each criterion produced with gain formula is shown on obtained by equation (4):
equation (2). The following plot will display, in a graph, the
result shown in table I into figure 1 that shows us the rank
bar of weighted criteria and weighted sub-criteria determined  ∑  
by the chance of gain for criteria and entropy for sub-criteria.
The value of Wj is gain value of each criteria. The greater
the value (Vi) is, the more preferred the alternative (Ai) is
and the faster the student completes the degree. The result of
degree-completion preference value by simple additive
weighting method shown in TABLE II

TABLE II. PREFERENCE VALUE OF SAW

No Student Number Preference Value (Vi)


1 1415015xxx 0.4177
2 1415015 xxx 0.4184
3 1415015 xxx 0.4622
4 1215015 xxx 0.4417
5 1115015 xxx 0.3788
6 1115015 xxx 0.4033
7 1115015 xxx 0.4081
8 1115015 xxx 0.3651
9 1115015 xxx 0.4287
10 1115015 xxx 0.4593

450 1007055 xxx 0.091

D. Weighted Product (WP)


Weighted product method process is similar to simple
additive weighting in setting the decision matrix (X) than
using multiplication to connect rating attributes, in which
each attribute rating should be raised first to the relevant
attribute weights. This process is similar to the process of
Fig. 1. Bar Chart of Criteria and Sub-Criteria Weight normalization [10].

Figure 1 points out the comparison of each sub-criterion Below is the formula to normalize the weight and obtain
influence in regard to how fast students complete their result of vector Si:

58
   The Wj in equation (9) is the weight of the criteria that in this
∑ paper using gain value in table I.

The weight Wj (j=1,2,3) of each criteria must be normalized


by dividing the ratio of the criteria in which gain value is  A+=(y1+, y2+,…, yn+); 
used.

 A-=(y1-, y2-,…, yn-); 


 ∏  

In equation (6), Si means normalization in i alternative that with


uses multiplication of connected rating attributes. Below is
the preference relative Vi given as the following equation (7):
{

  
∏ ( )
{
The greater the value (Vi) is, the more preferred the
alternative (Ai) is and the faster the student completes the The distance from the positive ideal solution and distance
degree. The result of degree-completion preference value by from the negative ideal solution formula are shown by
weighted product method is shown in the following TABLE equation (12) and (13) below:
III:

 √∑ ( )  
TABLE III. PREFERENCE VALUE OF WP

No Student Number Preference Value (Vi)


1 1415015xxx 0.00261
2 1415015 xxx 0.00261  √∑ ( )  
3 1415015 xxx 0.00292
4 1215015 xxx 0.00279 The last step in TOPSIS method is discovering the
5 1115015 xxx 0.00238 preference values using the following equation (14):
6 1115015 xxx 0.00252
7 1115015 xxx 0.00255   
8 1115015 xxx 0
9 1115015 xxx 0.0027 The greater the value (Vi) is, the more preferred the
10 1115015 xxx 0.0029 alternative (Ai) is and the faster the student completes the
degree. The result of degree-completion preference value by

TOPSIS method is shown in the following TABLE IV:


450 1007055 xxx 0

E. TOPSIS TABLE IV. PREFERENCE VALUE OF TOPSIS


Technique for order of preference by similarity to ideal
No Student Number Preference Value (Vi)
solution (TOPSIS) is based on the concept in which the best
alternative was selected. It is not only that having the shortest 1 1415015xxx 0.9404
distance from the positive ideal solution, but also that having 2 1415015 xxx 0.94041
the longest distance from the negative ideal solution [11]. 3 1415015 xxx 0.99987
TOPSIS requires performance rating of each alternative Ai 4 1215015 xxx 0.99703
on each normal criteria Ci, below is the equation (8):
5 1115015 xxx 0.90684
6 1115015 xxx 0.9355
   7 1115015 xxx 0.93635
√∑
8 1115015 xxx 0.87997
9 1115015 xxx 0.98455
Positive ideal solution A+ and the negative ideal solution
- 10 1115015 xxx 0.99942
A can be determined as the following equation (10) and (11),
and based on the normalization weighted rating yij as the

following equation (9): 450 1007055 xxx 0.00132

 yij=wjrij 

59
F. Confusion Matrix comparison result in table VI and calculation using equation
The confusion matrix is obtained by calculating the (15), the accuracy result of WP is 0.61333 or 61.3% and the
accuracy, sensitivity and specificity measures. The matrix precision is 0.52852 or 52.85%. From the comparison result
denotes classified samples as true, while others as false and in table VII and calculation using equation (15), the accuracy
misclassified [12]. In this paper the accuracy and the result of TOPSIS is 0.60667 or 60.7% and the precision is
precision of simple additive weighting, weighted product, 0.52673 or 52.67%. The accuracy and precision rate of the
and TOPSIS are calculated by using confusion matrix. The methods are shown in figure 2.
dataset used as standard is real degree-completion time of
graduate student. Below is the accuracy formula (15): The Percentage of Accuracy and Precision

62

   61

60

TP is true positive, TN is true negative, FP is false 59

positive, and FN is false negative. For the positive predictive 58


value or precision formula are given as the equation below: 57

56

55
  
54

53
In equation (16) PPV is positive predictive value. In this
paper there are three classes of degree-completion time 52

namely fast, moderate, and slow. Hence the precision of each 51

method is the average precision of each class. 50


SAW WP TOPSIS
In the academic dataset it is shown that 95 students are
Accuracy Precision
fast graduated, 98 students are moderate graduate, and 257
students are slow graduate as discovered from 450 dataset of
student degree-completion. The preference value in TABLE
II, III, and IV are divided into three groups as fast, moderate, Fig. 2. Bar Chart of The Percentage of Accuracy and Precision
and slow. Each group is then compared with the real degree-
completion time discovered in graduate’s data. The result of Figure 2 shows that WP has better accuracy and precision
comparison of each method is shown in the confusion matrix rate than other methods even though the difference is not too
below: significant in term of accuracy and precision rate.

IV. CONCLUSION
TABLE V. CONFUSION MATRIX OF SAW
From the work result and discussion, it can be concluded
SAW that undergraduate student degree-completion time can be
Fast Moderate Slow predicted with multi-criteria decision-making. Based on the
Fast 41 27 27 criteria weighted in this case study, weighted method (WP) is
Real
better to use than simple additive weighting (SAW) nor
Moderate 22 39 37
technique for order of preference by similarity to ideal
Slow 32 32 193 solution (TOPSIS) with 61.3% accuracy and 52.85%
precision. The weighting of ten criteria using entropy shows
TABLE VI. CONFUSION MATRIX OF WP us some interesting criteria having the biggest impact in the
degree-completion time. There are grade point average,
WP
student’s gender, and age when they registered. The result of
Fast Moderate Slow this research can be used to help academic staff in making
Fast 40 28 27 decision support system for student admission and
Real Moderate 26 39 33 administering mentoring process when the similar criteria to
Slow 29 31 197 this case study are available.

TABLE VII. CONFUSION MATRIX OF TOPSIS REFERENCES


[1] B. Prihasto, M. I. Irawan, and A. Masduqi, “Fuzzy MADM Method
TOPSIS for Decision Support System based on Artificial Neural Network to
Fast Moderate Slow Water Quality Assessment in Surabaya River,” J. Soft Comput. Decis.
Support Syst., vol. 1, no. 1, pp. 24–29, 2014.
Fast 40 26 29
[2] F. I. Perwitasari, A. A. Soebroto, and N. Hidayat, “Pemilihan
Real Moderate 23 40 35 Alternatif Simplisia Menggunakan Metode Weighted Product (WP)
Slow 32 32 193 Dan Metode Simple Additive Weighting (SAW),” J. Environ. Eng.
Sustain. Technol. JEEST, vol. 02, no. 01, pp. 20–30, 2015.
[3] H. Daniel Masethe and M. Masethe, Prediction of Heart Disease
From the comparison result in table V and calculation using Classification Algorithms, vol. 2. 2014.
using equation (15), the accuracy result of SAW is 0.60667
or 60.7% and the precision is 0.52684 or 52.68%. From the

60
[4] M. Tvaronavičienė and K. Razminienė, “Towards Competitive [9] M. Vujicic, M. Papic, and M. Blagojević, Comparative analysis of
Regional Development Through Clusters: Approaches To Their objective techniques for criteria weighing in two MCDM methods on
Performance Evaluation,” J. Compet., vol. 9, no. 4, pp. 133–147, example of an air conditioner selection, vol. 72. 2017.
2017. [10] P. Ciptayani and K. Dewi, Decision support system of e-book
[5] A. Setyawan, F. Yuni Arini, and I. Akhlis, Comparative Analysis of provider selection for library using Simple Additive Weighting, vol.
Simple Additive Weighting Method and Weighted Product Method to 953. 2018.
New Employee Recruitment Decision Support System (DSS) at PT. [11] D. Pratiwi, J. Putri, and D. Agushinta R., “Decision Support System
Warta Media Nusantara, vol. 4. 2017. to Majoring High School Student Using Simple Additive Weighting
[6] E. Triantaphyllou, Multi-Criteria Decision Making Methods: A Method,” Int. J. Comput. Trends Technol., vol. 10, pp. 153–159,
Comparative Study, vol. 44. 2000. 2014.
[7] M. Wati, Haeruddin, and W. Indrawan, “Predicting degree- [12] B. A. Alyoubi, “Decision Support System and Knowledge-based
completion time with data mining,” in 2017 3rd International Strategic Management,” Procedia Comput. Sci., vol. 65, pp. 278–284,
Conference on Science in Information Technology (ICSITech), 2017, Jan. 2015.
pp. 732–736.
[8] S. Singh and P. Gupta, “Comparative Study ID3, Cart and C4.5
Decision Tree Algorithm: A Survey,” Int. J. Adv. Inf. Sci. Technol.,
vol. 3, no. 7, pp. 47–52, 2014.

61

Potrebbero piacerti anche