EFSM-MLB: An Ensemble Feature Selection Model for Better Outcome Prediction in Major League Baseball Using Filter and Embedded Methods

Major League Baseball (MLB) stands as one of the most globally renowned and widely played tournaments at the international level in the realm of sports research. Predicting the key input variables of a match in MLB based tournament is very challenging. The selection process involves choosing which variables are more important for match prediction, as teams often use Sabermetrics in feature selection for an accurate selection of players. The current study aims to identify the major input variables that influence MLB team winnings. The authors of this research suggested an ensemble feature selection model for a better and more accurate outcome of a match. The proposed mechanism is tested on an open-access dataset of major leagues from 2005 to 2023, which is freely available on Baseball-Reference. The authors implement the proposed model on a set of sixty different offensive and defensive game features. Results obtained from deep analysis and implementation using linear regression and Correlation indicate a positive or negative association with win percentage. Here, the suggested model ranks all MLB variables from highly correlated to lesser correlated variables according to their association with win percentage. Pitching characteristics are found to be more important for forecasting match outcomes in favour of winners during this feature selection process. Furthermore, it has been discovered that run differential is a major factor in match prediction.