Calculating Win Probability in Korean Professional Baseball Using Deep Learning

This study proposes a method for providing viewers with richer and more personalized information in professional baseball broadcasts through new media platforms by leveraging big data and deep learning technologies.Specifically, the study aims to help viewers gain a more intuitive understanding of the game's progress by utilizing win probability metrics.In this research, a new model for calculating win probability is proposed by employing deep learning to learn the progress of baseball games.Notably, to address the issue of insufficient data for deep learning training, a new method was proposed that calculates win probability by utilizing expected runs, rather than relying on the traditional win probability calculation methods, and calculating win probability for each inning.As a result, it was confirmed that the new model trained through deep learning could predict outcomes at nearly the same level as traditional win probability calculation methods.Additionally, this method was implemented in actual internet broadcasts.Moving forward, by further developing this approach, it will be possible to provide viewers watching sports through new media with real-time, diverse information related to the game.This study used limited data from Korean professional baseball.With more games and data in the future, the model could become more accurate.