Team-Building Algorithm Based on Successful Records: Capstone Course

This report examines a successful cases-based (SC) algorithm for team building in capstone courses. The study, published in the Global Journal of Engineering Education, proposes an algorithm that utilizes successful team records from past capstone courses to assist students in forming effective teams. The algorithm calculates similarity measures between current virtual teams and past successful teams based on students' major subjects and GPAs. Through simulations, the SC algorithm, with an 85% matching rate, outperformed random team-building methods (38%). The report details the algorithm's procedure, including the calculation of average similarity coefficients and Euclidean distances, and its application using Microsoft Excel. It also compares the performance of the SC algorithm with random team building. The author believes that the accumulation of successful case information over time will make the SC algorithm even more effective for student team-building activities in the capstone course. The study highlights the importance of collaborative work and effective team formation in project-based learning within engineering education. The research includes a review of related algorithms such as clustering, fuzzy matching, max-sum, and skill parity methods, discussing their limitations and suitability for capstone course environments.

Volume 19, Number 3, 2017 © WIETE 2017
Global Journal of Engineering Education
243
INTRODUCTION
Project-based learning (PBL) is a naturally-occurring instructional method of learning process in which students draw
outputs with other team members after team building for cooperation. Students also use information to conduct reality
projects similar to those offered by a professor [1][2]. Students cooperate with co-workers to draw a method of solving
a problem and study the process of producing a common solution [3]. Further, in the PBL course, such as the capstone
course in an engineering school, students have to work as part of a team, and they can acquire experience in problem-
solving through cooperative communication. Whether the learning effect is positive or not depends on how students
cooperate with other team members [4][5].
In reality, at the initial stage of the capstone project, building an autonomous team is quite a high hurdle to confront,
compared with several other problems. For example, from the survey on students who had taken the capstone course in
the Department of Industrial Management Engineering at Hankuk University of Foreign Studies in Korea, over 74% of
students in 2015 and 80% in 2016 responded that the team-building process is one of the more difficult pieces of work
in the project execution. This study proposes a team-building algorithm to enable the carrying out of an effective PBL
for students taking the capstone course. In order to increase their project performance, the algorithm is based on the
member records of successful teams in past capstone courses. The successful cases-based (SC) algorithm, therefore,
may provide a similar team construction to the past successful teams.
The remainder of this article is comprised of three sections. The second section deals with the related algorithms,
focusing especially on methods matching members. In the third section, the author examines the details of the SC
algorithm and provides the simulation test to compare the SC algorithm with the current team-building method. In the
last section, he discusses future research and provides a conclusion.
RELATED ALGORITHMS FOR TEAM BUILDING IN ENGINEERING EDUCATION
Besides educational use, many firms in the industrial area have become interested in how to build an effective team for
industrial objectives. Thus, there have been various methods for building a team through matching suitable members,
such as clustering, fuzzy matching method, max-sum method and skill parity method. Those algorithms have been
applied in various fields including sports and job research, as well as in the education area. Clustering algorithms like
A team-building algorithm based on a successful record for a capstone course
Moon-Soo Kim
Hankuk University of Foreign Studies
Yongin City, Republic of Korea
ABSTRACT: The capstone course in engineering schools is one of the most important courses for students aiming to
become competent professional engineers capable of solving real industrial problems. Collaborative work and active
interaction are prerequisite for resolving the project. From the project outset, however, they confront the problem of
organising team members. This study proposes an algorithm based on the member information of successful team
records from the past capstone courses, which enables students to organise an autonomous team to carry out their project
successfully. The successful cases-based (SC) algorithm provides the most suitable candidates to be team members by
similarity measures between a currently virtual team and the past successful team based on the credit of major subjects
taken by students and their GPA. Through several simulation tests for the case of the 2016 capstone course of the
Department of Industrial Engineering, Hankuk University of Foreign Studies in Korea, the SC algorithm with 85% matching
rate showed better performance than the existing team-building method based on a random construction of teams from
among students, with 38%. The author believes that the accumulation effect of successful cases information over the years
will make the SC algorithm more effective for students’ team-building activity in the capstone course.
Keywords: Capstone design, project-based learning, team-building algorithm, successful cases-based algorithm, industrial
engineering

244
K-means clustering and agglomerative clustering have a function to bind objects with similar characteristics [6][7].
In team building, the diversity factor is important, because it allows groups to have a wide range of task knowledge,
skills and abilities. Also, it enables having a different perspective on a given task, so teams can discuss it more and seek
information to enhance the quality of their decision. Therefore, diversity can contribute to team performance [8][9].
Through a clustering algorithm, students are classified by their characteristics and also distributed to teams, to ensure the
diversity of the team. Furthermore, it enables a team member to be selected who is similar to a past team member for
a new team if students want. However, since the clustering algorithm requires various information about target students,
it has limitations in applying it to PBL courses.
Secondly, the fuzzy matching algorithm has usually been used on job searching sites by providing a degree of matching
of their requirements between job recruiters and seekers [10]. Although it is possible to find a team that can satisfy the
requirements of both sides, it may not be suitable for a number of students who participate in the PBL.
Thirdly, the max-sum strategy algorithm has been applied to form a team to maximise the sum of ability for positions to
form the best team [11]. The ability score of players is recorded in each position, and players are arranged in only one of
the positions by applying the rule maximising the sum of score of the team for building the best team [12]. Because the
algorithm selects team members based on best performance criteria, it recommends excellent students to build a team
when applied to the PBL course and to exclude the others. Thus, it may not be suitable for educational purposes.
The last related one is a skill-rarity algorithm that preferentially allocates the members who have a rare skill after
arranging the projects in order of value [13]. This method is a more powerful algorithm for a target team’s performance
than the performance of other teams in the PBL course.
Although the algorithms reviewed for building the team are useful in each area, the purpose and lack of information in
the PBL environment require the development of a new algorithm to help students form teams with high potential for
success in all their projects. The next section deals with a new algorithm using the information of the past PBL courses.
SUCCESSFUL CASES-BASED ALGORITHM
The Procedure of the Successful Cases-based (SC) Algorithm
The purpose of the SC algorithm is to support students in building teams similar to past cases (or teams) that obtained
high scores in their projects. Matching ratios with past successful cases is provided to teams that intend to recruit new
team members or students who want to participate in some teams, and they can also consider each other in regard to
team building through this matching ratio. The SC algorithm developed in this study calculates the matching ratio
between the team being composed and past successful cases by calculating the average similarity of a group process for
optimisation of clustering [14]. Figure 1 shows the SC algorithm procedure.
Figure 1: The SC algorithm procedure.
The algorithm begins with a notice from the recruiter who wants new members in the team and requests from applicants
who want to join the team. For example, when an applicant requests to be a member of the kth recruiting team in the
capstone course, the algorithm calculates the average similarity coefficient caik, based on their major courses taken
before the capstone course, for all team members in the kth recruiting team, as well as the average similarity coefficient
paik for all team members in every past successful team as shown in Figure 2 and Figure 3. The measures using

245
information on major courses are taken or not according to the fact that the project execution depends mainly on the
knowledge the team members have, and how well they understand and apply methodologies to solve their problems.
By using the two average similarity coefficients of the recruiting team and past successful teams, the algorithm calculates the
Euclidean distance dk between the recruiting team and one of the past successful teams that indicate the dissimilarity
between two teams. To normalise dk between 0 and 1, the algorithm sets simply it nk as 1/(1+ dk) that indicates approaching
to 1 if the recruiting team is becoming more similar to the past successful team, i.e. it can mean the recruiting team has
a high possibility of success. Lastly, the SC algorithm in Figure 1 calculates PMAT k using n k and the GPA k of the past
successful team k, i.e. PMATk = nk GPAk/4.5, where 4.5 is the perfect score of the capstone course.
Figure 2: The average similarity coefficients of the recruiting team.
Figure 3: The average similarity coefficients of the past successful team.
Figure 4: Experiment of SC algorithm by using the Microsoft Excel program.

246
To give an example of the SC algorithm, one applies it to Microsoft Excel using the information of actual team
construction and courses taken by students who participated in the capstone course from 2013 to 2016. Project managers
(recruiters) of project cases performed in 2016 remained on their original teams but, the other team members were
changed from all participants in 2016, and the team that obtained the highest PMAT was compared with past successful
cases from 2013 to 2015.
In reality, the team starts with one person who is the project manager and sequence numbers were provided to the
project managers. Also, to make up the count of real team members of each team in 2016, the data of past cases
(from 2013 to 2015) were divided into 3-member teams, 4-member teams and 5-member teams. As shown in Figure 4,
13 teams existed in 2016, and there were 32 teams from 2013 to 2015 consisting of seven 3-member teams, seven
4-member teams and eighteen 5-member teams.
Figure 5 is a typical example of the experiment of the SC algorithm and shows that team building is processed on the
basis of project manager (recruiter) Subject 1, who participated in the capstone course in 2016. The SC algorithm
recommends Subject 2, Subject 3 and Subject 4 as team members based on the highest PMAT values.
Figure 5: The highest PMAT value in the example.
Performance of SC Algorithm
To show the proposed SC algorithm’s performance, a comparative analysis for consistency with similar past cases in
terms of project outcome (e.g. success or failure) between the results from random team building and the results from
team building with the SC algorithm was performed as shown in Figure 6. The comparative analysis uses the 32 teams in
2013 to 2015 and 13 teams in 2016 of the previous experiment to provide an example of the subsection above. In the
test, the project managers of 13 project teams performing in 2016 remained on their original teams, but all the other
team members were changed in 2016, and the most similar team is found from the teams performed in 2013, 2014 and
2015 at the same time.
Figure 6: Comparison between the SC algorithm and the current team building.
After the most similar case is chosen, the PMAT value is calculated based on whether GPA of a similar case is high or
low. If the GPA of the similar case is high, the PMAT of the experimental (recruiting) teams will be high. Meanwhile,
the PMAT of the experimental teams will not be high. Because of these features, the SC algorithm can help students
build teams similar to the past successful cases.
vs.

247
On the other hand, in random team-building experiments, which represent the current team-building method, except for
project managers who were fixed, all other team members are assigned randomly to each team by using random number
generation. Hereafter, this random team-building method chooses the most similar past case based on the information of
courses taken by team members. Through the matching of the past team’s record, such as success or failure of capstone
course, over the ten random team-building experiments, 38% of the composed teams are matched with past successful
cases on average as shown in Figure 7.
Figure 7: Matching percent with the past success teams between teams building by random experiments (1st to10th trial)
and teams building by SC algorithm (11th trial).
While the team-building experiment with the SC algorithm is similar to the random team-building experiment, PMAT
figures are utilised during the process of changing team members. By the time the PMAT figure reaches over 0.75, team
members are changed continuously.
As a result of this experiment, 85% of composed teams, which means 11 teams out of 13 teams, are matched with
successful cases during the 2013 to 2015 period. Because the matching percentage with past successful cases of team
building by the SC algorithm is 47% higher than that of random team building, the SC algorithm seems to be superior to
the current method and can help students construct a team effectively to execute their project in the PBL course.
CONCLUSIONS
In this study, the SC algorithm is proposed for the purpose of building teams similar to past successful cases in order to
increase the performance of future team projects in engineering school capstone courses. As the SC algorithm measures
the matching ratio based on the information of courses taken by students, it can give a high successful feasibility in
capstone project execution to student teams. Furthermore, the comparison analysis between random team-building
experiments and team building by the SC algorithm indicated that team building by the SC algorithm is much more
similar to the previous successful performance than the existing method.
Because the proposed algorithm is basically dependent on the similarity of the past successful team features based on the
information of courses taken by students, different information from students can affect the similarity and the algorithm.
It means that the SC algorithm is extensible and applicable to the various group-based learning courses and cooperative
team-based projects in industry, while it is limited if there is insufficient past information regarding project teams.
Despite several limitations of the SC algorithm, the accumulation effect of information on successful team cases over the
years will make it more effective for students’ team-building activity in capstone course.
Other studies to utilise the algorithm to various PBL courses should be tested. First of all, an information system
working the algorithm, such as a team-building support system should be developed as a Web-based system regardless
of place and time. Second, in cases when there is no past course information, alternative algorithms should be
considered and developed. Lastly, a new team-building model including all participants, such as teachers, students and
interested industrial parties linked to the capstone project should be also considered.
ACKNOWLEDGEMENTS
This research was supported by the Basic Science Research Programme through the National Research Foundation of
Korea (NRF) funded by the Ministry of Science, ICT and future Planning (NRF-2017R1A2B4005858) and also by the
Hankuk University of Foreign Studies (HUFS) Research Fund.
REFERENCES
1. Laffey, J.M., Tupper, T., Musser, D. and Wedman, J., A computer-mediated support system for project-based
learning. Educational Technology Research and Develop., 46, 73-86 (1998).

248
2. Markham, T., Lamer, J. and Ravitz, J., Project Based Learning Handbook: a Guide to Standards-Focused Project
Based Learning for Middle and High School Teachers. (2nd Edn), Oakland, United States: Buck Institute for
Education (2003).
3. Howard, B.S., How to Design a Problem-Based Curriculum for the Preclinical Years. Springer Series on Medical
Education, New York, United States: Springer Pub Co. (1985).
4. Kim, M-S., A comparative review on problem- and project-based learning and applied method for engineering
education. J. of Engng. Educ. Research, 18, 65-76 (2015).
5. Kreijns, K., Kirschner, P.A. and Jochems, W., The sociability of computer-supported collaborative learning
environments. J. of Educational Technol, and Society, 5, 8-22 (2002).
6. Ivanovska, S., Ivanoska, I. and Kalajdziski, S., Algorithms for effective team building. Proc. 10th Conf. for Infor.
and Infor. Technol., 80-84 (2013).
7. Sahin, Y.G.., A team building model for software engineering courses term projects. Computer & Educ., 56, 916-
922 (2011).
8. Williams, K.Y. and O’Reilly, C.A., Demography and diversity in organizations: a review of 40 years of research.
Research in Organizational Behavior, 20, 77-140 (1998).
9. Polzer, J.T., Milton, L.P. and Swann, W.B., Capitalizing on diversity: interpersonal congruence in small work
groups. Administrative Science Quarterly, 47, 296-324 (2002).
10. Kim, H.R. and Jeong, I.S., Bi-directional fuzzy matching algorithm. J. of Korean Society for Internet Infor., 12, 69-
76 (2011).
11. Gale, D. and Shapley, L.S., College admissions and the stability of marriage. The American Mathematical
Monthly, 69, 9-15 (1962).
12. Kim, Y.S., Choi, S.B. and Han, T.Y., Construction algorithm for the strongest team beforehand game. The Korea J.
of Sports Science, 23, 735-743 (2014).
13. Kang, B.H., Kim, T.H., Bae, D.H., Lee, J.M. and Kim, S.W., An efficient team formation method for maximizing
profit in the expert network. The Korean Institute of Infor. Scientists and Engineers, 39, 89-91 (2012).
14. Yim, D.S. and Oh, H.S., Application of genetic and local optimization algorithms for object clustering problem
with similarity coefficients. J. of the Korean Institute of Industrial Engineers, 29, 90-99 (2003).
BIOGRAPHY
Moon-Soo Kim is a Professor of the Department of Industrial and Management Engineering,
at Hankuk University of Foreign Studies (HUFS) in Korea. He gained considerable
experience as a project investigator at the Electronics and Telecommunications Research
Institute (ETRI) in Korea, prior to joining the University. His research focuses on
technology and service management and its various application fields, as well as recently
engineering education concerning student-centred learning theory and practice. He has
published papers in several international journals, such as the International Journal of
Engineering Education, Technology Analysis and Strategic Management, Omega, ETRI J.,
Tele. Policy, Telemactics and Informatics, Scientometrics, and Technological Forecasting
and Social Change, Service Science, International Business Review, etc, and in several
domestic journals.