Identification of soft clusters of pitch traces and its application to predict pitcher-versus-batter matchups

Department of Industrial EngineeringIn this study, we cluster the trace data of the pitch in a way that connects two clustering algorithms. And we propose a new feature that represents the characteristics of pitchers through quantitative and quali tative analysis of soft clusters created by this method. In addition, based on the features created in this way, we provide examples of our methods being used by attempting predictions for matchups between pitchers and batters and predictions in a finer range. The predictions attempted through our clustering re sults show more reliable performance for outlier samples compared to conventional methods, and ensure good performance even in situations where the number of dataope