Binus University Repository

Denny Prabowo, Yulius

Spits Warnars, Harco Leslie Hendric

Lumban Gaol, Ford

Abdurachman, Edi

Soewito, Benfano

English

An initial research on Halstead's technique for pattern similarity relationship study

TANUAR, EVAWATY

Saleh Abbas, Bahtiar

Trisetyarso, Agung

Kang, Chul-Ho

Suparta, Wayan

Back Propagation Neural Network Experiment on 
Team Matchmaking MOBA game 
1,2Evawaty Tanuar 
1Computer Science Department, 
School of Computer Science, 
2BINUS Graduate Program – Doctor 
of Computer Science, 
Bina Nusantara University 
Jakarta, Indonesia 11480 
etanuar@binus.edu  
 
Bahtiar Saleh Abbas 
Computer Science Department, 
BINUS Graduate Program – Doctor 
of Computer Science, 
Bina Nusantara University 
Jakarta, Indonesia 11480 
bahtiars@binus.edu  
 
 
Agung Trisetyarso  
Computer Science Department, 
BINUS Graduate Program – Doctor 
of Computer Science, 
Bina Nusantara University 
Jakarta, Indonesia 11480 
atrisetyarso@binus.edu 
 
  
Chul-Ho Kang  
Computer Science Department, 
BINUS Graduate Program – Doctor 
of Computer Science, 
Bina Nusantara University 
Jakarta, Indonesia 11480 
chkang5136@kw.ac.kr 
 
 
 
 
Ford Lumban Gaol  
Computer Science Department, 
BINUS Graduate Program – Doctor 
of Computer Science, 
Bina Nusantara University 
Jakarta, Indonesia 11480 
fgaol@binus.edu 
 
 
 
 
2,3Wayan Suparta 
2BINUS Graduate Program – Doctor 
of Computer Science, 
Bina Nusantara University 
Jakarta, Indonesia 11480 
3Department of Electrical 
Engineering, 
 Sanata Dharma University 
Yogyakarta, Indonesia 55282 
drwaynesparta@gmail.com 
 
 
Abstract—Backpropagation is one of the famous method for 
learning, Implementing Backpropagation techniques in games is 
part of AI in game. This paper aim to analyzed the best team 
composition in a Moba Game type in learning using neural 
network. Mobile Legend: Bang Bang is one of MOBA game for 
mobile phone. In this experiment, the 2 hidden layers back-
propagation neural network algorithm is used to analyze the 
Mobile Legend game.  5000 training steps was done to reach the 
stability. This also a way of learning to find the best composition 
in a team played. Research in area AI and game is very wide and 
open, a huge opportunity to be explore in future. The objective of 
this research is to produce the good combination of team which 
can be resulted in a greater winning chance.   
Keywords—back propagation; neural network; AI learning; 
game; matchmaking 
I.  INTRODUCTION 
Neural network is one of the techniques for learning. Since 
it first introduced in 1943 by McCulloch and Pitts, neural 
network was used in several different areas for learning and 
prediction. A lot of research been done using this method, 
range from predicting financial, tourism demand and a lot of 
simulation tools was built to support research using neural 
network for learning and prediction.  
Most of the developer and player focus on the graphical and 
playable features. The learning and decision side somehow is 
ignored. The challenges in AI in game are the time allocation 
for developing AI rather that graphics, the games always 
change over the time[1]. The rich of data and environment in 
game is interested to be learn. Learn about the model also can 
increase the knowledge and the chance to win a game, also can 
be implemented as part of the Artificial Intelligence. 
II. BACKGROUND 
A. Neural Network 
Neural network also known as parallel distributed 
processing, neural computation, it models the brain activity as 
the basic idea. Inspired by Neuroscience, it contains network of 
brain cells call neuron.  It can calculate corresponding desire 
input to targeted output[2]. 
 
Fig. 1. Model of a Neuron (Russel & Norvey)  
Method in Neural Network usually used to know as Feed 
Forward Method. The process starts from the beginning 
continue to the next neuron until the output. The formula used 
are show is the fig 1[3]. First layer usually is the Input Layer, 
 
2018 International Conference on Information and Communications Technology (ICOIACT)
978-1-5386-0954-5/18/$31.00 ©2018 IEEE 240
continue with the Hidden Layers (minimal 1 layer), and the last 
layer is the Output Layer or for result. Each layer at least 
contains one or more node and neuron to connect the node. 
B. Back Propagation Algoritm 
Back Propagation Algorithm follow the feed forward 
network rules, which have been improved with two main 
characteristics. The first is the activation function of the hidden 
unit, and the second is the gradient of the activation 
function[4]. 
In short, the neural networks learning procedures are: 
a. Calculate the values or output units, observed the 
difference between computed and expected output 
values 
b. Start from output layer, moved to hidden layer, 
propagate and update the weight between the two 
layers. 
Repeat those steps until the desired result or network 
converges, means that the difference between the expected and 
the forecast output are in an acceptable range. So, basically it 
consist of two phases, a forward and a backward pass[5]. 
Input vectors corresponding to target vector also can be 
added with bias. Bias is a sigmoid layer are able for 
approximating the function. There are different methods in 
deciding the number of neurons, try and error method, rule of 
thumb, simple method, two phase method and sequential 
orthogonal approach[6]. From the experiments, can be 
concluded that if the number of hidden layers increase, the 
accuracy also increases, but ne network became more complex. 
 
C. MOBA Game – Mobile Legend 
MOBA – Multiplayer Online Battle Arena is a sub-genre of 
the classic Real Time Strategy (RTS) type game. The different 
is that in this type of gameplay, there is no base construction 
and players only control on one character[8]. Two teams are 
fighting to destroy the base of opponent which is at the end of 
the maps. 
The most famous game in this genre is DotA (Defense of 
the Ancients), by Valve, or Warcraft III: Reign of Chaos, The 
Frozen Throne, and lately there is Mobile Legend: Bang Bang 
in Android platform. This genre is the type that eSport 
tournament very famous about lately. This game mostly played 
by 2 teams, where each team consists of 5 players with 
different type of hero. It this why that team matchmaking is 
important in this type of game. 
 In Mobile Legend, there are 53 heroes that categorized in 6 
types: tank, fighter, assassin, mage, marksman, and support. 
Also, there are 11 heroes that categorized between 2 types.  
 
 
Fig. 2. Some of the heroes in Mobile legend 
 
Fig. 3. Maps of Mobile legend (www.mobilelegends.com) 
Typically, the maps of MOBA games shows in fig 3. Two 
different compete to destroy the opponents. 
III. METHODOLOGY 
In this research, experiment on Mobile Legend game is 
conducted to analyze the effectiveness composition of team in 
the game. The type of the battle are classic and ranked only. 
56 team compositions gathered from ranked and classic 
type competition to be used as data training and testing. Using 
Neural applet Version 4.3.8, written by Kevin O’neill,et.all 
from University of British at Columbia[9].  
Input is the 6 different types of the heroes, calculated from 
the type of heroes used by the team, 1 for each hero. For those 
heroes that in between type, 0.5 was given for each type. For 
winning team, score 1 for the output, and for the losing team, 
the output was 0. 
Several tests been done. This application using the back-
propagation learning algorithms. 
IV. EXPERIMENT 
A. Experiment Data 
To be able to do learning and testing, the data is important. 
56 data from 28 games was retrieved from different player with 
the random possibility method where from player in classic and 
ranked genre.  From the data collected, each type in the team 
are scored as the definition of character by mobilelegend.com. 
The average scores for each type of hero that frequently used 
are: 
 
x Tank: 0.519231 
x Fighter: 1.2211 
 
2018 International Conference on Information and Communications Technology (ICOIACT)
241
x Assassin: 0.9807 
x Mage: 0.84615 
x Marksman: 1.3365 
x Support: 0.076 
x Winning: 0.5 
 
It means that most of the player prefer more on fighter and 
marksman type hero rather than other. 20 % of the data being 
used as testing data in this learning experiment. 
B. Experiment Process 
This research will analyze the 6 – 6 – 3 – 1 formation. 6 
input nodes, 2 hidden layers with 6 nodes in the first layer and 
3 nodes for the second layers, and 1 node as the result. The 
result is for predicting the winning or losing possibility. The 
first one is without bias and second experiment with bias. 
These two figures below show the random initialization for 
2 different experiments. Fig 4 is the neural network 
initialization without Bias node and in Fig 5, neural network 
with Bias node. 
 
 
Fig. 4. Neural Network Initialization 
 
Fig. 5. Neural Network With Bias Initialization 
 
C. Experiment Result 
For each experiment, 5000 times steps training have been 
done. The experiment result shows the predictive correction 
can reach to 90% in the without Bias model compare to with 
Bias model only reach 86%.   
TABLE I.  5000 TRAINING RESULT 
Result Without Bias With Bias 
Training Error 2.3445 3.8542 
Test Error 1.2604 1.9656 
Predictive Correctly 90% 86% 
Predictive Incorrectly 10% 14% 
 
Even that the prediction training error and test error is 
better from the without Bias model, but the Bias model more 
stable after the 1000 training compare to without Bias. Shows 
in the figure below. 
 
 
Fig. 6. Result comparation in grafic 
D. Team Matching Result 
Used the best backpropagation learning result, 90% on 
experiment 1, the base result calculated as below.  
 
 
 
 
 
TABLE II.  BASE RESULT FOR TYPE IN A TEAM 
 
 
 
2018 International Conference on Information and Communications Technology (ICOIACT)
242
Tank Fighter Assassin Mage MM Support Result 
5 0 0 0 0 0 0.99840 
0 5 0 0 0 0 0.00100 
0 0 5 0 0 0 0.00070 
0 0 0 5 0 0 0.01160 
0 0 0 0 5 0 0.86230 
0 0 0 0 0 5 0.00000 
 
The type of heroes that influence much in the team are tank 
and marksman type. Meaning that a team full of support 
definitely will lose. 
126 combinations created from 6 types of heroes with 
maximum score 3 for each type. The results are categorized as 
in table III. Only 23 combinations out of 126 results in 
possibility larger than 0.99% winning chance. Also, only 31 
combinations will have a chance to win. 
TABLE III.  POSSIBILITY TEAM COMBINATION  
BP Result Total Result Percentage 
> 0.9 23 18.25% 
> 0.8 28 22.22% 
> 0.5 31 24.60% 
0 13 10.32% 
 
TABLE IV.  BEST TEAM COMBINATION 
Tank Fighter Assassin Mage MM Support Result 
2 1 1 1 0 0 1 
1 2 1 0 1 0 0.9996 
0 1 2 1 1 0 0.9997 
3 1 0 0 1 0 0.9998 
 
The best team combination as the result from this 
experiment can be seen in table IV.  Another conclusion are: 
x Support more than 1 in a team will lose 
x 2 tanks in a team will have a bigger chance to win 
V. CONCLUSION 
This research successfully tested a back-propagation 
algorithm on the artificial neural network which can be used to 
predict the best team composition for the Mobile Legend: Bang 
Bang - MOBA game. From the experiments results, shows that 
only 24.6% of combination have more that 50% chance to win. 
It shows that how important to know the team combination as 
part of a strategy to win a game. In addition, it is necessary to 
collect more data and be analyzed to understand the 
components capabilities in order to increase the chances of 
winning.  
From this research can proceed to model the best 
compositions heroes compete with other formations. This 
information will capable to improve AI's ability to have a 
greater chance of winning a game. Also in further research, 
more data can be gathered and analyzed to have a better 
understanding the component that able to improve the winning 
chance. It is true that human or the player ability also impact in 
the game, but AI in game will only learn from something that 
can be calculated. 
AI in game, neural network, learning and decision making 
have a huge possibility to be a research area in future. 
 
REFERENCES 
[1] A. S, S. L. H, and P. Gonz, “Game AI for a Turn-based Strategy Game 
with Plan Adaptation and Ontology-based retrieval,” Proceeding 
DIMEA ’08 Proc. 3rd Int. Conf. Digit. Interact. Media Entertain. Arts, 
no. 642882, pp. 295–302, 2008. 
[2] A. D. Dongare, R. R. Kharde, and A. D. Kachare, “Introduction to 
Artificial Neural Network,” Int. J. Eng. Innov. Technol., vol. 2, no. 1, 
pp. 189–194, 2012. 
[3] P. Russel, Stuart; Norvey, Artificial intelligence: a modern approach. 
2009. 
[4] R. Law, “Back-propagation learning in improving the accuracy of neural 
network-based tourism demand forecasting,” Tour. Manag., vol. 21, pp. 
331–340, 2000. 
[5] A. Goyal, G. K. Walia, and S. Kaur, “IMPLEMENTATION OF BACK 
PROPAGATION ALGORITHM,” Int. J. Inf. Technol. Knowl. Manag., 
vol. 5, no. 2, pp. 429–431, 2012. 
[6] F. S. Panchal and M. Panchal, “Review on Methods of Selecting 
Number of Hidden Nodes in Artificial Neural Network,” Int. J. Comput. 
Sci. Mob. Comput., vol. 311, no. 11, pp. 455–464, 2014. 
[7] S. V Kolwankar, “Evolutionary Artificial Intelligence for MOBA / 
Action-RTS Games using Genetic Algorithms,” Int. J. Comput. Appl., 
pp. 29–31, 2012. 
[8]  https://www.mobilelegends.com/ 
[9] http://aispace.org 
2018 International Conference on Information and Communications Technology (ICOIACT)
243


Back propagation neural network experiment on team matchmaking MOBA game

Akun, Akun

Lembar Peer Review "Ecocriticism Values In The Indonesian Folktale Timun Emas (Golden Cucumber)"

Chowanda, Andry

 
LEMBAR  
HASIL PENILAIAN SEJAWAT SEBIDANG ATAU PEER REVIEW 
KARYA ILMIAH : PROSIDING* 
 
Judul Makalah : Forming of Dyadic Conversation Dataset for Bahasa Indonesia 
Penulis Makalah : Andry Chowanda, S.Kom., M.M., Ph.D., MBCS., CCP 
Jumlah Penulis : 3 orang 
Identitas Pengusul : Penulis pertama/penulis ke-3/penulis korespondensi ** 
Identitas Prosiding :  a. Judul Prosiding : 3rd International Conference on Computer Science and 
Computational Intelligence, ICCSCI 2018 
  b. ISBN/ISSN : 1877-0509 
  c. Tahun Terbit, Tempat Pelaksanaan : 7-8 September 2018, Tangerang, Indonesia 
 d. Penerbit/organize : Elsevier B.V.  
 e. Alamat repository PT/web prosiding  :  
 https://www.sciencedirect.com/science/article/pii/S1877050918314686/pdf?md5=3
7bd1b34e6ad5915299431e5a0f1aeae&pid=1-s2.0-S1877050918314686-main.pdf 
  f. Terindeks di (jika ada)   : Scopus 
 
Kategori Publikasi Makalah 
(beri √ pada kategori yang tepat) 
 Prosiding Forum Ilmiah Internasional Terindeks 
Prosiding Forum Ilmiah Internasional 
Prosiding Forum Nasional 
 
Hasil Penilaian Peer Review : 
  
 
Komponen Yang Dinilai 
Nilai Maksimal Prosiding  
Nilai Akhir 
Yang 
Diperoleh 
Internasional 
Terindeks 
Internasional Nasional 
a. Kelengkapan unsur isi paper (10%) 0.6   0.51 
b. Ruang lingkup dan kedalaman   
    pembahasan (30%) 
1.8   1.53 
c. Kecukupan dan kemutakhiran   
    data/informasi dan metodologi (30%) 
1.8   1.53 
d. Kelengkapan unsur dan kualitas terbitan/  
    prosiding (30%) 
1.8   1.53 
Total = (100%) 6   5.1 
Nilai Pengusul =  
 
Jakarta, 04-11-2020 
 
 
 
 
 
Dr. Derwin Suhartono, S.Kom., M.T.I. 
Unit kerja: Universitas Bina Nusantara 
 
*Dinilai oleh dua Reviewer secara terpisah 
**Coret yang tidak perlu 
 
 
 
 
 
 
 
 
  
 
√ 
 
 
√ 
 
 
 
LEMBAR  
HASIL PENILAIAN SEJAWAT SEBIDANG ATAU PEER REVIEW 
KARYA ILMIAH : PROSIDING* 
 
Judul Makalah : Forming of Dyadic Conversation Dataset for Bahasa Indonesia 
Penulis Makalah : Andry Chowanda, S.Kom., M.M., Ph.D., MBCS., CCP 
Jumlah Penulis : 3 orang 
Identitas Pengusul : Penulis pertama/penulis ke-3/penulis korespondensi ** 
Identitas Prosiding :  a. Judul Prosiding : 3rd International Conference on Computer Science and 
Computational Intelligence, ICCSCI 2018 
  b. ISBN/ISSN : 1877-0509 
  c. Tahun Terbit, Tempat Pelaksanaan : 7-8 September 2018, Tangerang, Indonesia 
 d. Penerbit/organize : Elsevier B.V.  
 e. Alamat repository PT/web prosiding  :  
 https://www.sciencedirect.com/science/article/pii/S1877050918314686/pdf?md5=3
7bd1b34e6ad5915299431e5a0f1aeae&pid=1-s2.0-S1877050918314686-main.pdf 
  f. Terindeks di (jika ada)   : Scopus 
 
Kategori Publikasi Makalah 
(beri √ pada kategori yang tepat) 
 Prosiding Forum Ilmiah Internasional Terindeks 
Prosiding Forum Ilmiah Internasional 
Prosiding Forum Nasional 
 
Hasil Penilaian Peer Review : 
  
 
Komponen Yang Dinilai 
Nilai Maksimal Prosiding  
Nilai Akhir 
Yang 
Diperoleh 
Internasional 
Terindeks 
Internasional Nasional 
a. Kelengkapan unsur isi paper (10%) 0.6   0.48 
b. Ruang lingkup dan kedalaman   
    pembahasan (30%) 
1.8   1.44 
c. Kecukupan dan kemutakhiran   
    data/informasi dan metodologi (30%) 
1.8   1.44 
d. Kelengkapan unsur dan kualitas terbitan/  
    prosiding (30%) 
1.8   1.44 
Total = (100%) 6   4.8 
Nilai Pengusul =  
 
Jakarta, 04-11-2020 
 
 
 
 
 
Dr. Fredy Purnomo, S.Kom., M.Kom. 
Unit kerja: Universitas Bina Nusantara 
 
*Dinilai oleh dua Reviewer secara terpisah 
**Coret yang tidak perlu 
 
 
 
 
 
 
 
 
  
 
√ 
 
 
√ 
 
 


Form Reviewer "Forming of Dyadic Conversation Dataset for Bahasa Indonesia"

Girsang, Abba Suganda

Check 36. Proceeding NICTE-
3.pdf
by Arta Moro Sundjaja
Submission date: 08-Nov-2018 09:24AM (UTC+0700)
Submission ID: 624024364
File name: 36._Proceeding_NICTE-3.pdf  (442.49K)
Word count: 3737
Character count: 20960








89%
SIMILARITY INDEX
3%
INTERNET SOURCES
89%
PUBLICATIONS
4%
STUDENT PAPERS
1 89%
Exclude quotes On
Exclude bibliography On
Exclude matches < 1%
Check 36. Proceeding NICTE-3.pdf
ORIGINALITY REPORT
PRIMARY SOURCES
Ahmad Nurul Fajar, Frits Septian Johanes,
Abba Suganda Girsang, Gunawan Wang.
"Evaluation of the used of visual electronic
banking system application using UTAUT 2 in
Bank Central Asia corporation", IOP
Conference Series: Materials Science and
Engineering, 2018
Publicat ion


Similarity "Evaluation of the used of visual electronic banking system application using UTAUT 2 in Bank Central Asia corporation"

Sasmoko, Sasmoko

Check JJA Prof Sasmoko _
Scopus 2018 - 14 -
Neuroresearch Another form of
mixed method.pdf
by Arta Moro Sundjaja
Submission date: 10-Aug-2018 09:25AM (UTC+0700)
Submission ID: 624024728
File name: copus_2018_-_14_-_Neuroresearch_Another_f orm_of _mixed_method.pdf  (2.05M)
Word count: 5540
Character count: 32376



















29%
SIMILARITY INDEX
26%
INTERNET SOURCES
9%
PUBLICATIONS
4%
STUDENT PAPERS
1 21%
2 3%
3 2%
4 1%
5 1%
Check JJA Prof Sasmoko _ Scopus 2018 - 14 - Neuroresearch
Another form of mixed method.pdf
ORIGINALITY REPORT
PRIMARY SOURCES
www.sciencepubco.com
Internet  Source
Sasmoko, F Doringin, Y Indrianti, A M Goni, P
Ruliana. "Indonesian Teacher Engagement
Index (ITEI): An Emerging Concept of Teacher
Engagement in Indonesia", IOP Conference
Series: Materials Science and Engineering,
2018
Publicat ion
irep.iium.edu.my
Internet  Source
uav.ro
Internet  Source
Sigit Wijaksono, Sasmoko, Y Indrianti, SA
Widhoyoko. "Jakarta socio-cultural ecology: a
sustainable architecture concept in urban
neighbourhood", IOP Conference Series: Earth
and Environmental Science, 2017
Publicat ion
6 1%
Exclude quotes On
Exclude bibliography On
Exclude matches < 1%
Submitted to Universiti Teknologi MARA
Student  Paper


Similarity "Neuroresearch: Another form of mixed method"

Tanty, Heruna

Check Linear regression model
using.pdf
by Arta Moro Sundjaja
Submission date: 31-Oct-2018 10:06AM (UTC+0700)
Submission ID: 619532227
File name: Linear_regression_model_using.pdf  (491.84K)
Word count: 3445
Character count: 18308







89%
SIMILARITY INDEX
4%
INTERNET SOURCES
89%
PUBLICATIONS
5%
STUDENT PAPERS
1 89%
2 1%
Exclude quotes On
Exclude bibliography On
Exclude matches < 1%
Check Linear regression model using.pdf
ORIGINALITY REPORT
PRIMARY SOURCES
Syarifah Diana Permai, Heruna Tanty.
"Linear regression model using bayesian
approach for energy performance of
residential building", Procedia Computer
Science, 2018
Publicat ion
Submitted to Coventry University
Student  Paper


Similarity "Linear regression model using bayesian approach for energy performance of residential building"

Moro Sundjaja, Arta

Form Reviewer "The adoption of Facebook as internet marketing strategies in journal promotion"

Form Reviewer "Review assessment support in Open Journal System using TextRank"

Oktavia, Tanty

Peer Review "Analyzing MOOC features for enhancing students learning satisfaction"

Binus University

Not a member yet