Domain-Independent Sports Match Prediction using Dynamically Updated Database Open Access

Xu, Chenxi (Fall 2020)

Permanent URL: https://etd.library.emory.edu/concern/etds/pr76f462t?locale=en
Published

Abstract

This thesis presents the design of sports match predictions with various approaches, including linear models based on different evaluation metrics of players' performance and neural networks. Our main goal is to incorporate this prediction model into the modern conversational AI and help the bot to make more interesting and human-like conversations. Therefore, our models have high interpretability to convey information about various aspects of a sports match. While our model can talk about players' performance of past matches, predictions of further matches and players' impact, it can still achieve an accuracy of 68.7% on the NBA and 67.5% on the MLB match predictions. Moreover, our model has a high generality and applicable to new sports. Hence, it will be easy for further development and expansion after being incorporated into the conversation system.

Table of Contents

1 Introduction 1

2 Background 4

2.1 Sports Match Outcome Prediction . . . . . . . . . . . . . . . . 4

2.1.1 Sports Performance Indicators . . . . . . . . . . . . . . 4

2.1.2 Statistical Model . . . . . . . . . . . . . . . . . . . . . 5

2.1.3 Neural Network Approach . . . . . . . . . . . . . . . . 6

2.2 Conversational AI . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2.1 Alexa Prize . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Approach 12

3.1 Database Structure . . . . . . . . . . . . . . . . . . . . . . . . 12

3.1.1 Dynamically Update . . . . . . . . . . . . . . . . . . . 14

3.2 Baseline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.3 Ensemble Learning Method . . . . . . . . . . . . . . . . . . . 15

3.3.1 Performance Score . . . . . . . . . . . . . . . . . . . . 16

3.3.2 Future Performance Prediction . . . . . . . . . . . . . 20

3.3.3 Match Result Prediction . . . . . . . . . . . . . . . . . 21

3.4 Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Experiments 25

4.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.2 Regularization Term . . . . . . . . . . . . . . . . . . . . . . . 26

4.3 Experimental Result . . . . . . . . . . . . . . . . . . . . . . . 27

4.3.1 Impact of Covid-19 . . . . . . . . . . . . . . . . . . . . 28

4.3.2 Approaches Comparison . . . . . . . . . . . . . . . . . 29

5 Analysis and Interpretation 31

5.1 Player Performance Analysis . . . . . . . . . . . . . . . . . . . 31

5.1.1 Box Score Category Weight . . . . . . . . . . . . . . . 31

5.1.2 Practical Usage . . . . . . . . . . . . . . . . . . . . . . 33

5.2 Player and Team Impact . . . . . . . . . . . . . . . . . . . . . 34

5.3 Possible Conversation . . . . . . . . . . . . . . . . . . . . . . . 36

6 Conclusion 39

Bibliography 40

About this Honors Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files