[2Win5] Poster session 2

[2Win5-84] Development of a Half-Inning Prediction Model Based on Baseball Game Progress

Shodai Usami¹, 〇Akito Nakano¹, Yuki Iwata¹, Ryunosuke Onishi¹, Shota Shiiku¹, Jun Ichikawa¹ (1.Shizuoka University)

Keywords:baseball, time-series prediction, LSTM, machine learning, skill

This study proposes a half-inning prediction model for batting outcomes in baseball, such as a hit, using time-series data. In our model, which employs a Long Short-Term Memory (LSTM) network, uniquely categorized past batting outcomes within a game are input, and the frequencies of batting outcomes in the next half-inning are predicted. The results showed that the proposed model tended to predict both frequent and rare outcomes by weighted Mean Squared Error (MSE) as the loss function. Meanwhile, it suggests that new loss functions and evaluation indices need to be introduced in this model for higher accurate predictions. Our findings may provide a first step for accurate prediction using complex time-series data in sports.