An Approach to Predict Outcomes in Sports Games with Bigdata Techniques and Data Mash-Up

Abstract

The purpose of this study is to predict performance of players and teams in baseball games using data mining and data mash-up. In this paper, decision tree technique and data mash-up approach are used for predictions. A data set on 111 games by one of the most outstanding baseball player in S. Korea, player A(S.Y. Lee) and his team is collected for the study. Three sets of data are combined for the mash-up from 3 different sources: Korea Baseball Organization(KBO), Korea Meteorological Administration, and Google Trends. The results from the analysis have 3 findings. Firstly, the important variables for ‘H(Hits)’ are google trends and humidity. If ‘trend’ is 25 and more and ‘humid’ is 77.1 or over, the probability to make one or more hits is 85.7%. It is most likely for player A to make one or more hits when public interest is high and the weather is humid. Secondly, the number of spectators and humidity are significant for ‘BB(base on balls)’. If ‘spect’ is 12823.5 or more, ‘humd’ is 62.2 or above and ‘humd’ is below 73.2, probability to get ‘BB’ is 57.1%. When there are many spectators and it is moderately humid, the probability for getting ‘base on balls(walk)’ is high. Thirdly, wind speed and temperature are important to have a good ‘result’. If ‘wind’ is 2.75 or over and ‘temp’ is 25.1 or above, probability to get a team victory is 100%. When the wind blows a little and the temperature is high, his team will win. This study focuses on a baseball team and a player. Further study can extend the scope of applications to other teams and other sports.