Volume 11 Number 4
World Institute
for
Engineering
and Technology
Education (WIETE)
World Transactions on
Engineering and Technology
Education
MELBOURNE 2013
Editor-in-Chief:
Zenon
J.
Pudlowski
Wor ld Transactions on Engineering and Technology Education
Z.J. Pudlowski
G.C. Kabouridis
C-H. Huang
A.H. Al-Arfaj & H.I. Abu-Mulaweh
Z.Ke
H. Deng & X. Song
Z. Ned ic, S. Gadzhanov & A. Nafalski
P. Anastasiadis & G. Metaxas
L. Furst & V. Mahnic
X. Liu, Y. Sun & J. Wu
J. Qiu
AP. Wibawa, A. Nafalski &Z. Nedic
A. Jurcius & AV. Valiulis
W. Cao, Z. Liu & L. Zheng
K.W.H. Tsui & S.J. Thatcher
D. Schott
X. Lv
S. Gadzhanov, A. Nafalski &Z. Nedic
S.J. Thatcher & K.W.H. Tsui
J. Kang & G. Chen
W.Li
X. Liu
Co nte nts
Editorial 359
The niche institute strategy - the way out of economic crisis for Greek higher 360
educational institutions: the case of the Technological Educational Institute of Western Greece
A competency-based calculus teaching module for engineering students 365
A newly established college of engineering led by a non-engineer 37 I
Engineering education knowledge management based on Topic Maps 376
The theory and practice of linear regression 382
An industry-based project for teaching electrical power systems 388
Formulating the principles of an eco-city 394
Introductory programming course: motivating students with prior knowledge 400
Reforming the teaching of mechanical design for industrial design students 406
Evaluating students' skills in a physical education course using grey system 4 12 theory
Potential of an educational application of a language statistical translation 418
system
Searching for residual stress measurement methods for structural steel 424
components
Reforming the teaching of a single chip microprocessor course based on 428
CDIO engineering education
Aviation safety in the undergraduate curriculum: aviation students' 434
perceptions
Signal reconstruction - a project for students of electrical engineering 438
Peer assessment of perception and attitudes in public speaking English classes 445
Remote laboratories on motion control systems 450
A comparison of the learning outcomes of international students from Asia 456
and Australian students in an undergraduate aviation programme
Computer-aided instruction of mechanical engineering test technology and 460
signal processing based on the use of MATLAB
A study on matching the criteria of core courses with job requirements: a 465 case study based on the urban rail specialisation in vocational colleges
An evaluation index system of undergraduate education quality based on an 470
S. Jia & C. Yang
W. Cai, W. Gu, L. Zhu, W. Lv
& S. Dong
D. W.S. Tai, J-L. Chen,
R-C. Zhang & V . Tai
H. Zhang, Y. Lu, J. Yang, G. Xu
&Z. Sun
F. Jia
B. Cai
H. Feng & Y. Li
L. Zhang, W.Zhang,N. Zhu & Y. Jia
S. Kocijancic
& A. Boonsongsrikul
J. Kasih, M. Ayub & S. Susanto
S. Li, C. Yang & Y. Zhang
H. Huang, S. Chen & C. Huang
P. Yao, C. Zhang, R. Yao, Y. Xiao & X. Wang
C.Xu
H. Liu
J-R. Wang, Y-C. Chang, C-W. Li & H-C. Li
L. Wang, X. Jin, Y. Shi & L. Cao
X. Zhang, Z. Xiong, Y. Ding
&G . Wang
J. Wang, C. Ma, S. Huang, L. Yu &R.Xia
D. Chen & Y. Wang
Q. Lei &Z. Wu
Y. Wang, H. Liu & D. Chen
X. Wu & L. Shang
Index of Authors
Teaching software testing based on CDIO セ@
6
Training engineering undergraduates in building environment and n fg\' 480
The review of current engineering and technology programme accreditation 484
in Taiwan
Applying Bus mode to the teaching of computer basics and the use of 490
competitions in teaching
A multi-agent collaborative learning scheme for young university teachers 495
based on reinforcement learning
Engineering education of electrical information courses: a comparison 500
between Germany and China
An effective teaching method for a Linux course 504
Reforming cryptography education by teaching practical sessions 508
A survey of student-centred approaches to engineering education - a case 513
study concerning Slovenia and Thailand
Predicting students' final passing results using the Apriori Algorithm 517
A survey on the present situation of students' educational technology ability 521
at a normal college
Computer teaching based on experiments using a virtualisation platform 527
Analysing and improving teaching based on a random sample investigation 532
Classroom flipping as the basis of a teaching model for the course Mobile 537
Application Development
Constructing a Web-based autonomous learning model for teaching English 541
A study of competency indicators of Web-based micro-entrepreneurs in the 547
consumer leisure and entertainment electronics retail industry, using AHP for analysing the weights
An evaluation of the quality of university education based on multidimensional 553
time series analysis
Reforms of an operating system course at a local university 559
Applying Whitehead's process philosophy to engineering-oriented practical 565
education
Reconstructing the training system of garment engineering education with 569
the characteristics of art engineering
Research on engineering applications of Fresnel reflection functions 574
Establishing a training mode for art engineering using art classification codes 579
Adding value to college business majors through a model education station 584
built for practice sessions
589
Th
、・セ@
COi
ed1 Gr
Ar
En
Th
Er.
mz
SUI
pn en
re:
n
ecWorld Transactions on Engineering and Technology Education Vol.II. No.4, 2013
©20 13 W/ETE
Predicting students' final passing results using the Apriori Algorithm
Julianti Kasih t , Mewati Ayubt
&
Sani
sオウ。ョエッセ@Maranatha Christian University, Bandung, Indonesiai· Parahyangan Catholic University, Bandung, Indonesia;
ABSTRACT: The studies discussed form part of a progranune, other aspects of wh ich have been previously considered [1)[2]. The ultimate objective is to facilitate a lecturer in helping students to predict their final passing results based on their performance in several subjects in the first four semesters of their study period. In previous research, this aim was achieved through two techniques: discriminant analysis [I) and the Classification and Regression Trees (CART) algorithm [2). Those two techniques resulted in a diagramme-based relationship. In this research, a rule-based relationship of the form
IF - THEN
is introduced and subsequentl y applied using software based on the Apriori Algorithm .INTRODUCTIO
This research project continues a theme that was considered in previous stud ies [1)[2) , the objective of which was to facilitate a lecturer in helping students to predict their final passing results based on their performance in several subjects in the first four semesters during their study period. The arguments for why this kind of prediction is considered important were discussed in [2) and are rewritten in the Appendix. The passing results in the Indonesian education system are classified into three grades: Extraordinary (Cum Laude), Very Satisfactory and Satisfactory [3].
The research was undertaken in the same institution, the Faculty of Information Technology, a university in Bandung, West Java, Indonesia. For reasons of confidentiality, the full name of the institution has not been included. In the two previous works, it was demonstrated that discriminant analysis [I) and the Classification and Regression Trees (CART) algorithm [2) helped academic advisors in this faculty to predict the final passing results of a student based on his/her grade in some subjects during the first four semesters during their undergraduate progranune. This sort of facility enables academic advisors to ass ist students in setting up their study plans each semester in order for them to per form to their full potential [1)[2). Moreover, this work aims at helping the academic advisors with a more practical way of predicting the final passing res ults of a student.
In this research, a data mining task called an association was employed. Association is performed through a technique called the Apriori Algorithm. This algorithm produces some rule-based relationships in the form
IF- THEN
statements. This kind of statement serves in a more ready to read feature compared to the territorial map or decision tree employed in the previous work in [I) and [2) , respectively.OVERVIEW OF BACKGROUND THEORY
David Hand et al define data mining as the analysis of (often large) observational data sets to find unsuspected
relationships and to summarize the data in novel ways that are both understandable and useful to the data owner [4].
The observational data or the data to be summarised are often called the training data. Data mining has six tasks: description, estimation, prediction, classification, clustering and association [4]. Association is based on affinity analysis, the study of attributes or characteristics that go together. One amongst several methods for affinity analysis is market basket analys is, which tries to discover associations among these attributes with the aim to discover association
rules for quantifying the relationship between two or more attributes.
The association rule takes the form
If
antecedent, then consequent, which for reasons of simplicity often desires a single consequent [4). The performance measures of this rule are the support, confidence, rule support, lift and deployabilityterm instances. Instances define the number of records in the data set that match the antecedents. For example, given the association
If
purchase bread, then purchase cheese, the number of records in the training data that include the antecedent purchase bread are referred to as instances.Support or antecedent support is the proportion of training data for which the antecedents are true. For example, if 50% of the training data includes the purchase of bread, then, the rule
If
purchase bread, then, purchase cheese will have an antecedent support of 50% . Support as defined here is the same as the instances but is represented as a percentage.Rule support is the proportion of training data for which the entire rule, antecedents and consequent(s), are true . For example, if 20% of the training data contains both the purchase of bread and cheese, then, rule support for the rule
If
purchase bread, then, purchase cheese is 20%.
Confidence is the ratio of rule support to antecedent support. This indicates the proportion of training data with the specified antecedent(s) for which the consequent(s) is/are also true. For example, if 50% of the training data contains bread (indicating antecedent support) but only 20% contains both bread and cheese (indicating rule support) , then, confidence for the rule
If
purchase bread, then, purchase cheese would be rule support/antecedent support or, in this case, 40%. Lift is the ratio of confidence for the rule to the prior probability of having the consequent. For example, ifI 0% of the entire population purchases bread, then, a rule that predicts whether people will purchase bread with 20% confidence will have a lift of20/IO = 2. If another rule tells that people will purchase bread with 11% confidence, then, the rule has a lift of close to one (I), meaning that having the antecedent(s) does not make a lot of difference in the probability of having the consequent. In general, rules with lift far from I will be more interesting, than, rules with lift close to one(!).
Deployability is a measure of what percentage of the training data satisfies the conditions of the antecedent but does not satisfy the consequent. In product purchase terms, it basically means what percentage of the total customer base owns (or has purchased) the antecedent(s) but has not yet purchased the consequent.
EXPERlMENT: THE RESULT AND INTERPRETATION
The research was undertaken in the same institution, the Faculty of Information Technology, a university in Bandung, West Java, Indonesia. The rules were generated by SPSS Clementine I 0.1 software.
As in the previous research programme, the academic transcripts from 146 alumni served as input or observational or training data, which were available from the authors [ I ][2]. From these data, the students ' final passing resulls were determined by the final marks from the following eight (8) subjects: IF I 02 (Introduction to Computer App lication), IFI03 (Introduction to Information Technology), IF104 (Algorithms and Programming), IF105 (Basic Programming), IF I 06 (Informatics Mathematics) and IF202 (Linear Algebra and Matrices), IF 203 (Computer Network) and
IF
205 (File System and Access). The final marks of these eight (8) subjects take the role of antecedents. The final marks ofa subject were classified into five (5) groups, as follows: A (High Distinction), B (Distinction), C (Credit), D (Pass) and E (Fail) with some intermediates, such as B+ and C+.The students ' final passing results, as mentioned previously, were classified into three groups: I - Extraordinary (Cum Laude); 2 - Very Satisfactory; and 3 - Satisfactory. The students ' final passing results take the role as the consequent.
- --- - --
-F]、NウエセBG M セャZイQPNQ@
ANセ@ f&t !"Ud セ@ !o:>i .. JGBBBGセ@ セ@ セ@
i セ@ tllif '(!I · • ·l> " r'
w
to > <> •*
o. o. •.-' ... r.J
®-®-<§)
t
ᄋ セ MMMMMMMMMMMMMMMMMMMMMMMM MMMᄋᄋ ᄋ セ@..
o セ エ セエ ᄋ セ N oN ゥ@ . JWIO.S, N Pイ セ l ᄋセ@ • o....a
Figure I: The classification model representation in Clementine I 0.1 .
, r ·
-」[hZ Cdエゥj ェ セ@ 。 セ セ@
ᄋ M セセB@セ ヲ^。セ wB a@
.
[[LNN,,__
セLN⦅セ@,,,_
Z[L[L セ@Th
•
•
Thi
The
The SPSS Clementine 10. 1 model is described in Figure I. In thi s fi gure, the icon on:
• the left side describes the input or observational or training data;
the right side describes the algorithm employed, in this case, the Apriori Algorithm ; and the top-middle side describes the output of the Apriori Algorithm ;
the bottom-middle side describes the type of the data, the final marks of the 8 (eight) subjects are the antecedents, while the pass ing result is the consequent.
The training data were saved in the form ofa SPSS worksheet file, a section of which is represented in Figure 2.
37-m1A
37:;IDSA
373X6A 372003A 37XXBA
372013 B+
3720U A
372015 c 37Xll6>.
SWRPセW@ A 372018 A
372022 A
372023 A
37201JA
37JJ25 A
37JJ26 B+
37m7 A 370ID3 A 372034 B 37:Jr£ A
37Xl38 A
3721)!DA
SWセWa@
1FI02 IF103
A
A
B
.c
',;
. !f lO!
A B• A A 0 C• A A A 8 • A c C• A 8 A 0 B+ A c c A B• c A A A s-A A A A A A A A A A A A A A A A A .A c
JF2Q!?__l _ セ@ L M セ ョ ⦅ウゥ ョァセ ョオセ M J ⦅ZZ⦅ セ@___ ___i _ _ ><-•• _L_ .:'¥. - ·-··
B• Very Sa11sfattory
s-8
Very Satisfactory
VerySat isfaclory
Vr1yS;m;faclory
E d1ao:dinary
e セ エQ。 ッQ、 Q ョ 。イケ@
VerySa1 isfac tory _Very Satisfa ctory ___ _
セ b @
VerySatisfaclory vn;-satisfaclory Ezl rao:dina ry
Very Saiisfaclory
V e1yS<11isf.ictory
Very Satisfactory y erySa1isfactory Very Satisfactory _Ertraordlnary VerySam.faclory A :e c c A .A 8 8+ 8 A B• B·
..
8+ C+ a+ B• B• 8 • s [ゥエ ゥ ウヲ。」エッセ@Very Satisfactory
eセ QQ。ッイ、ゥョ。イケ@
Very Sat isfa ctory Very Satisfactory
VeryS a11sfae1ory
Figure 2: Some part of the training data.
The generated association rules are displayed in Figure 3. In generating association rules, SPSS Clementine I 0. 1 software gives the user the options to determine the minimum antecedent support, the minimum rule confidence and the maximum number of antecedents, which in this research were set at 20%, 80% and 5, respectively. Six association rules were generated. The authors will be able to generate more rules if they reduce the minimum antecedent support and the minimum rule confidence and vice versa.
- ·
Consequent Ante cedentI
Instances I Supporl%M v[NョsTQヲゥ^ヲセヲャクNエ@ lf105 = C\i. 30 Ig UセX@
PassingResun = Very Salisfaclory IF103 = B
IF106 = A 30 20.5£8
lf102 = A Passin!;;Resun = Very Sati sfactory IF103 = 8
lf102 = A
..
30.137= VerySatisfaclory IF103 = B 56 38.356
=Very Satisfactory lf1 03 = 8 39 25.712 lf106.= A
PassingResun= v・イケs。エゥウヲ \Q 」エッセ@ ャヲQセT ] X@ 38 26.027
セ@ 」ッイ N エ。・ョ」・セ@ Rule Support % I
i9fl.000 18,193
90.000 18.193
88.636 26.71 2
87 .500 33.562
87 .179 23.289
84 .211 21.918
U1\ 1.3V 1.327 1.307 1.290 1.286 1.242
EJ o1 :• :
f n onl...,.hil""
2.055.
2.055
3.,25
U 95
3 . .t25
4.1 10
ᄋ M セセMMセセセMMセᄋMBセᄋᄋMMNBセセMMM セBM ᄋMᄋMMM Zセ@ .11!
セセBゥB l セセゥエゥョッセセMM[NL M[イMMM]M ᄋMM ᄋ MM GBGMMM MM BM ᄋMᄋMM MMMMMBMMBGMMMMM ᄋ@---·· ---=---==:--=
w
t_canca1_J
I
セーセ@ jc セ Mセ@•
•
•
•
One of the rules generated is IF IFl03 =Band IF/06 =A and IFI02 =A, THEN Passing Resu/1 = Very Satisfactory.
The antecedents of this rule is IFI03 =Band IF/06 =A and IFl02 =A , and its consequent is Passing Result= Ve1y
Satisfactory. This rule has the following performances:
• Instances equal to 30, which means that out of 146 records in the training data sets, the number of records in the data set that match the antecedents is 30;
Support or antecedent support is 20.548%, this value is due to the fact that the number of records in the 146 training data for which the antecedents are true is 30, or 20.548%;
• Rule support is 18.493%, this value is due to the fact that the number of records in the training data for which the antecedents and consequent(s) are true is 27 out of 146 records or 18.493%;
• Confidence is 90%, which means that the ratio of rule support, that is I 8.493%, to antecedent support, that is 20.548% is 90%;
Lift is 1.327, this value comes from the ratio of confidence for the rule (90%) to the prior probability of having the consequent Passing Result
=
Very Satisfactory (99 records out of I 46, or 67.8%);• Deployability is 2.055%, this value comes from a measure of what percentage of the training data satisfies the conditions of the antecedent IF/03
=
B and IF/06=
A and IF/02=
A, but does not satisfy the consequentPassing Result = Ve1y Satisfactory. Jn this case is three (3) records out of I 46 or 2.055%.
Care should be taken when applying the generated rules. Those rules do not express the causal relationship between the antecedent(s) and the consequent. As discussed in the background theory, association is based on affinity analysis, the study of attributes that go together. In the association rule context, the attributes are the antecedent(s) and consequent.
CONCLUSIONS AND SUGGESTION FOR FURTHER RESEARCH
This research demonstrates the ability of one of the data mining techniques that enable an academic advisor to help students predict their final passing results . Compared to the previous research [1][2], this research demonstrates that the prediction can be performed very practically, with neither graphs nor charts being required. The prediction is carried out by sentences of the form ,
If
antecedent, then consequent. Further research may take the form of the investigation of the prediction of the final passing results, which might be based on a multivariate statistics technique called logisticregression. The authors intend to perform this research in the not too distant future .
ACKNOWLEDGMENT
Sincerest thanks are conveyed by the authors to Mr Radiant Victor Imbar, former Dean of the Faculty of Information Technology at Maranatha Christian University, for his support and providing the data from the academic transcripts of alumni.
REFERENCES
I. Kasih, J. , and Susanto, S., Predicting students ' final results through discriminant analysis. World Transactions on
Engng. and Techno/. Educ., JO, 2, 144-147 (2012).
2. Kasih, J., Ayub, M. and Susanto, S., Predicting students' final passing results using the Classification and Regression Trees (CART) algorithm. World Transactions on Engng. and Techno/. Educ. , 11 , I, 46-49 (2013). 3. Government Regulation, the Republic of Indonesia, Nr 60 Year I 999 about Higher Education.
4. Larose, D.T., Discovering Knowledge in Data: An Introduction to Data Min ing. Hoboken, New Jersey, USA: John Wiley & Sons, Inc. (2005).
APPENDIX
In relation to the Introduction, the arguments for making the prediction of students' final passing results as the mai n issue of this research are as follows:
•
First, one of the important aims of higher education in the Republic of Indonesia is to prepare the academicparticipants (students) to become members of society with the academic
and/or professional ab1liiies to enable
them to apply/develop/enrich the foundations of knowledge in the sciences, technology and the arts.
Second, to achieve this aim, undergraduate students are assigned to several academic advisors (an informal translation of the Indonesian term Dasen Wali) throughout their years of higher educational studies.
Third, academic advisors, who are lecturers, have as their main task the fostering of students ' academic and non-academic activities . With regard to students ' non-academic activities, one of the duties of the non-academic advisor is to help students in setting up their study plans for each semester.
Fourth, setting up a study plan includes providing guidance for students regarding how many subjects, and whi subjects, to undertake.
Fifth, through this guidance, students are expected to obtain the best passing results at the end of their undergraduate study [1-3].
World Transactions on Engineering and Technology Education
/\ssociat ion ofTu iwan Engineering
Ed uca ti on and Man age me nt (/\ T EEM ) Ta ipei. Taiwa n
K L Univers ity Vijayawada. Gun tu r District
An dh ra Pradesh, India
Univers ity of Bots" ana
Gaborone. Bots\\ ana
Publi shed by:
Col lege of Engineeri ng &
Eng.im.-cring T echnology
Northern Jll inois U ni vers ity
DeK alb, Il linois. "Si\
NYU=prJly
POLYTECHNIC INSTITUTE OF NYU
Polytechnic Institute o f' New York niversity
New Yo rk. SA
Co llege o f Ma nageme nt
H ungk uang ni vcrsity Ta ichung. Ta iwan
Ru ssian A ssociat ion of Engineering Un iversi ti es
Moscow. Russi a
World Institute fo r Eng ineering and Technology Educat ion (WI ET E) 34 Hampshire Road, Glen Waverley, Melbourne, VIC 3 150, A ustralia
E-mai l: info@wiete.com.au Internet: ht tp://www.wiete.com.au
c 201 3 Wl ETE
!SS 1446-22 57
Comm onwea ll h Sc ience and Techn ology Academy for
Researc h (C-ST/\R) Chennai. India
Technol ogical Educat ion Instit ute of Piraeus
Pi raeus- A thens. Greece
Art icles published in the World Transact ions on Engineering and Techno logy Education (WT E&T E) have undergone a fo rmal and rigorous process of peer rev iew by in te rnational referees.
Thi s Jo urnal is copyright. Apart fro m any fair dealing for the purpose of private srudy, research, criticism or review as permitted und er th e Copyright Act, no part o f th is Journal may be reproduced by any process witho ut the written perm iss ion of the publisher.
Responsibility jar 1he conlen/s of lhese arlicles resls upon !he au/hors and 1101 !he publisher. Dala presen!ed and