[논문]Quantitative Analysis for Win/Loss Prediction of 'League of Legends' Utilizing the Deep Neural Network System through Big Data

No, Si-Jae; Moon, Yoo-Jin; Hwang, Young-Ho

doi:10.9708/jksci.2021.26.04.213

[국내논문] Quantitative Analysis for Win/Loss Prediction of 'League of Legends' Utilizing the Deep Neural Network System through Big Data 원문보기

韓國컴퓨터情報學會論文誌 = Journal of the Korea Society of Computer and Information, v.26 no.4, 2021년, pp.213 - 221

No, Si-Jae (Dept. of Comp. Engr. & African Languages, Hankuk University of Foreign Studies) , Moon, Yoo-Jin (Dept. of Mgmt. Information System, Hankuk University of Foreign Studies) , Hwang, Young-Ho (Division of Public Admin. & Economics, Kunsan National University)

초록
AI-Helper

이 논문은 League of Legends (LOL) 게임의 승패를 예측하기 위하여 Deep Neural Network Model 시스템을 제안한다. 이 모델은 다양한 LOL 빅데이터를 활용하여 TensorFlow 의 Keras에 의하여 설계하였다. 연구 방법으로 한국 서버의 챌린저 리그에서 행해진 약 26000 경기 데이터 셋을 분석하여, 경기 도중 데이터를 수집하여 그 중에서 드래곤 처치 수, 챔피언 레벨, 정령, 타워 처치 수가 게임 결과에 유의미한 영향을 끼치는 것을 확인하였다. 이 모델은 Sigmoid, ReLu 와 Logcosh 함수를 사용했을 때 더 높은 정확도를 얻을 수 있었다. 실제 LOL의 프로 게임 16경기를 예측한 결과 93.75%의 정확도를 도출했다. 게임 평균시간이 34분인 것을 고려하였을 때, 게임 중반 15분 정도에 게임의 승패를 예측할 수 있음이 증명되었다. 본 논문에서 설계한 이 프로그램은 전 세계 E-sports 프로리그의 활성화, 승패예측과 프로팀의 유용한 훈련지표로 활용 가능하다고 사료된다.

Abstract ▼ AI-Helper

In this paper, we suggest the Deep Neural Network Model System for predicting results of the match of 'League of Legends (LOL).' The model utilized approximately 26,000 matches of the LOL game and Keras of Tensorflow. It performed an accuracy of 93.75% without overfitting disadvantage in predicting the '2020 League of Legends Worlds Championship' utilizing the real data in the middle of the game. It employed functions of Sigmoid, Relu and Logcosh, for better performance. The experiments found that the four variables largely affected the accuracy of predicting the match --- 'Dragon Gap', 'Level Gap', 'Blue Rift Heralds', and 'Tower Kills Gap,' and ordinary users can also use the model to help develop game strategies by focusing on four elements. Furthermore, the model can be applied to predicting the match of E-sports professional leagues around the world and to the useful training indicators for professional teams, contributing to vitalization of E-sports.

주제어

표/그림 (8)

표 Table 1. Result of Evaluation of Variables
표 Table 2. Result of Data Normalization
표 Table 3. Result of Comparison between Softmax and Sigmoid Accuracy Function
표 Table 4. Comparison of the Activation Functions
그림 Fig. 1. Architecture of DNN
그림 Fig. 2. Accuracy and Loss Graph of the Model
표 Table 5. Loss Function Comparison
표 Table 6. Predictions of 2020 League of Legends Worlds Championship

AI 본문요약
AI-Helper

* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.

가설 설정

Therefore, our research aims to predict the outcome of the match with data such as ‘Minion Kills’, ‘Champion Kills’, ‘Wards Installation’, ‘Dragon Kills’, and ‘Rift Herald Kills’. In this regard, we hypothesized that the ‘Gold gap’ and ‘Level gap’ with the opponent team in the middle of the game would have the greatest impact on the match’s Win/Loss.

제안 방법

And then it constructed a Deep Neural Network model that predicted the Win/Loss results of LOL by entering the selected feature values processed by normalization of the input data. Finally, we performed evaluation of the model.
To measure the weight of each variable for the suggested model, it built a Single-Layered Neural Network with data such as ‘Gold Gap’, ‘Level Gap’, ‘Minion Kills Gap’, ‘Kill Gap’, ‘Ward Gap’, ‘Blue First Tower’, ‘Tower Kills Gap’, ‘Dragon Gap’, and ‘Blue Rift Heralds’ in 15 minutes. For the comparison of the feature selection model, the research also built a Multi-Layered Neural Network and evaluated the result and found the same result as the Single-Layered Neural Network. For the feature selection processing, the learning rate used was 0.
The dependent variable used on the dataset was ‘Blue team’s Win or Loss’. The independent variables used on the dataset were ‘Gold Acquisitions’, ‘Level Sum’, ‘Minions Kills’, ‘Ward Installations’, ‘Enemy champion Kills’, ‘Destruction of the first tower’, and ‘Dragon kills’ through processing of the feature selection.
To confirm the practicality of our program, the suggested model predicted 16 games of the '2020 Worlds Championship'. The model divided the test set into two, creating a validation set that further tests the final model after the first test. The suggested model used all the data of the Challenger rank games for learning in this model.
The purpose of this paper is to ensure that the match prediction of the LOL can be implemented through Deep Neural Networks using open data provided by Riot Games, not by the team history record but by the personal game processing record.
The research constructed the deep neural network model for prediction of the LOL match, utilizing Keras of TensorFlow [20, 21, 22, 23, 4]. It compared two sets of the game data up to 10 minutes and 15 minutes after the game had started.
The research found by experiments that ordinary users can also use the model to help develop game strategies by focusing on four elements (Dragon, Level, Rift Heralds, Tower). It is also expected to utilize the model for predicting match results of similar genre games.
It was difficult to secure objectivity because the characteristics of the champion vary from season to season, and the control and the teamwork differ greatly in the player’s personal ability. Therefore, our research aims to predict the outcome of the match with data such as ‘Minion Kills’, ‘Champion Kills’, ‘Wards Installation’, ‘Dragon Kills’, and ‘Rift Herald Kills&rsquo;. In this regard, we hypothesized that the ‘Gold gap’ and ‘Level gap’ with the opponent team in the middle of the game would have the greatest impact on the match’s Win/Loss.
This research utilized approximately 26, 000 matches of LOL game to perform feature selection of the neural network structure in order to select input variables. And then it constructed a Deep Neural Network model that predicted the Win/Loss results of LOL by entering the selected feature values processed by normalization of the input data.
Experiments showed that among the three activation functions, ReLU has the highest accuracy. Thus, the research used ReLU as the activation function of the model.
To decide the Hidden-Layer’s activation function, the research experimentally compared ReLU and Softmax and Tanh Function. Table 4 showed the result of this comparison.
To measure the weight of each variable for the suggested model, it built a Single-Layered Neural Network with data such as ‘Gold Gap’, ‘Level Gap’, ‘Minion Kills Gap’, ‘Kill Gap’, ‘Ward Gap’, ‘Blue First Tower’, ‘Tower Kills Gap’, ‘Dragon Gap’, and ‘Blue Rift Heralds’ in 15 minutes. For the comparison of the feature selection model, the research also built a Multi-Layered Neural Network and evaluated the result and found the same result as the Single-Layered Neural Network.

이론/모형

Keras and Python etc. The model focuses on an end-to-end approach to develop supervised learning algorithms in regression and classification with practical business-centric use-cases, which is often implemented in Keras provided by TensorFlow [9].
The research selected the K-fold method for the process of progressing the learning. The final accuracy value was 0.
The suggested model used ‘Gaussian Normalization’ as the normalization method [19]. ‘Gaussian Normalization’ is a method of normalizing x’ = (x – means) / standard deviation instead of the input x.
This research used a DNN model in Figure 1, consisting of four variables used for learning as input layers, two hidden layers with 64 nodes. It tried to identify how many hidden layers and nodes would make the highest prediction performance by adding 2 hidden layers and 16 nodes to the previous every time, and found that no big difference existed in performance between DNN with 2 hidden layers and 64 nodes and that with the more hidden layers and nodes.
8005) is clear. Thus, the research model is constructed using the Sigmoid function.

성능/효과

Furthermore, the prior work developed a system to predict LOL Win/Loss in real-time. However, our work focuses on making more accurate predictions by making LOL Win/Loss predictions after 15 minutes of the game, which is around half of the average playing game time in progress.
It tried to identify how many hidden layers and nodes would make the highest prediction performance by adding 2 hidden layers and 16 nodes to the previous every time, and found that no big difference existed in performance between DNN with 2 hidden layers and 64 nodes and that with the more hidden layers and nodes. Besides, it used the ReLU function as activation functions as hidden layers.
Overall, the key factor in winning and losing the game is the choice of champions for each position, the control of each player, the cooperation of team members, the growth through opponent minions, the destroy enemy turrets, the installation of wards, and the Dragon and Rift Herald. It was difficult to secure objectivity because the characteristics of the champion vary from season to season, and the control and the teamwork differ greatly in the player’s personal ability. Therefore, our research aims to predict the outcome of the match with data such as ‘Minion Kills’, ‘Champion Kills’, ‘Wards Installation’, ‘Dragon Kills’, and ‘Rift Herald Kills’.

후속연구

The future work to do is as follows. First, the research should continue to divide the dataset into beginner users, ordinary users, and top users at the experiments for the general purpose of the game result prediction. With the proper dataset, we can expect to gain results that are more accurate than prior experiments.
In this case, all the users of the game can use our model to predict their game personally. Second, the method provided in the research should be applied to the match prediction method for other E-sports professional leagues.

참고문헌 (23)

Yong Chen, Hong Chen, Anjee Gorkhali, Yang Lu, Liqian Ma, and Ling Li, "Big Data Analytics and Big Data Science: A Survey," Journal of Management Analytics, Vol. 3, No. 1, pp. 1-42, 2016.

상세보기
Teradata, "Big Data Analytics - Reveal the Best Opportunities for Big Data Companies," 2021. https://kr.teradata.com/Solutions/Big-Data?utm_campaign2020brand&utm_sourcegoogle&utm_mediumpaidsearch&utm_contentTERA1_GS_Brand_APAC-KR-EN_CV_NBKW_BMM&utm_creativeBigData-DataAtTheCenter|BigDataOverview&utm_termbig%20data%20science&gclidEAIaIQobChMIwLiR4o6C7wIV2KqWCh0TEw13EAAYASAAEgLGOvD_BwE
Sangho Kim, "A Study on Relationship of BDBA (Big Data Business Analytics) System and Supply Chain Management," Journal of Korea Research Association of International Commerce, Vol. 19, No. 2, pp. 89-107, 2019.

상세보기
Hyejeong Park, Kyoungha Seok, Juyong Shim and Changha Hwang, "Deep Learning from TensorFlow," Hanbit Academy Press, 2019.
Bruce Lehrman, "Big Data's Role in the Post-COVID Era," Data Agility, Vol. 16, Issue 11, Sept. 2020. https://www.pipelinepub.com.
Katy Warr, "Strengthening Deep Neural Networks: Making AI Less Susceptible to Adversarial Trickery," O'Reilly Media, 2019.
Wesley Chai, Mark Labbe, and Craig Stedman, "Big Data Analytics," 2021. https://searchbusinessanalytics.techtarget.com/definition/big-data-analytics
Jun Wu, Jian Wang, Stephen Nicholas, Elizabeth Maitland, and Qiuyan Fan, "Application of Big Data Technology for COVID-19 Prevention and Control in China: Lessons and Recommendations," Journal of Medical Internet Research, Vol. 22, No. 10: e21980, Oct. 2020. DOI: 10.2196/21980.

상세보기
Jojo Moolayil, "Learn Keras for Deep Neural Networks: A Fast-Track Approach to Modern Deep Learning with Python," Apress, 2019.
Jen-Tzung Chien, "Source Separation and Machine Learning," Elsevier: Academic Press, 2019.
Hon-Ki kim, Yu-Seop Kim. "League of Legends Win/Loss Prediction Using TensorFlow," Hallym University, 2017.
Senpai.gg. https://senpai.gg
Honqmei Li, Jinying Huang, and Shuwei Ji, "Bearing Fault Diagnosis with a Feature Fusion Method Based on An Ensemble Convolutional Neural Network and Deep Neural Network," Sensors (Basel, Switzerland), Vol. 19, Issue 9, pp. 2034, 2019.

상세보기
N. Yuvaraj, R. Arshath Raja, N.V. Kousik, Prashant Johri, and Mario Jose Divan, "Chapter15 - Analysis on the Prediction of Central Line-Associated Bloodstream Infections (CLABSI) Using Deep Neural Network Classification," Computational Intelligence and Its Applications in Healthcare, Academic Press, pp. 229-244, 2020. https://doi.org/10.1016/B978-0-12-820604-1.00016-9
Gilbert Lim, Wynne Hsu, Mong Li Lee, Daniel Shu Wei Ting, and Tien Yin Wong, "Chapter 21 - Technical and Clinical Challenges of A.I. in Retinal Image Analysis," Computational Retinal Image Analysis::Tools, Applications and Perspectives, Academic Press, pp. 445-466, 2019. https://doi.org/10.1016/B978-0-08-102816-2.00022-8
Xiangrui Xu, Yaqin Li, and Cao Yuan, "Identity Bracelets for Deep Neural Networks," IEEE Access, Vol. 8, pp. 102065-102074, 2020. DOI: 10.1109/ACCESS.2020.2998784

상세보기
Wonil Lee, Byungjai Kim, and HyunWook Park, "Quantification of Intravoxel Incoherent Motion with Optimized B-values Using Deep Neural Network," Magnetic Resonance in Medicine, Feb. 2021. DOI: 10.1002/mrm.28708

상세보기
2020 Challenger Rank, https://developer.riotgames
Junmo Cho, "Big Data Analytics and Artificial Intelligence Starting with Python," Infinity Books, 2020.
Michael Negenevitsky, "Artificial Intelligence," Addison-Wesley, 2011.
Russell, Stuart, "Artificial Intelligence: A Modern Approach," Pearson Education, 2017.
Ian Goodfellow, Yoshua Bengio and Aaron Courville, "Deep Learning," MIT Express, 2016.
Ingook Cheon, "Artificial Intelligence: Machine Learning and Deep Learning by Python," Infinity Books, 2020.

표제어: PCR

동의어: Packet Collision Rate

용어 설명 출처 목록 (6)

용어 설명: PCR은 세균 특이성이 있는 primer를 이용하여 적은 수의 세균이 있을지라도 쉽게 검출할 수 있는 유용한 방법이며, 이를 이용하여 구강 내 치면세균막이나 타액에서 직접 세균을 검출할 수 있게 되었다[8].

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증