Multimodal Machine Learning for Major League Baseball Playoff Prediction

Abstract

The introduction on sabermetrics has changed the way Major League Baseball (MLB) teams valued their players. Since then, new baseball stats have been made to make various predictions for MLB teams. With the immense amount of data on baseball players, teams, and scores. Using various Supervised machine learning algorithms, we plan to see how well we can accurately predict which teams will make it to the playoff for year 2019. For this research, we have gathered data from the last 20 years. The features that we will utilize for our machine learning algorithm includes Runs, Batting Average, Homeruns, Strikeouts, Innings Pitched, Earned Runs, and Earned Runs average. We decided to use a Logistic Regression model and a Support Vector Classifier (SVC) as the two machine learning algorithms for our features. After running our tests, our models showed that our trained algorithms were only able to predict accurately 77% of the teams correctly. Of those 77% accurately predicted, 59% was recalled correctly. This led to our overall projected model being only 60% accurate. As the projected model was only able to correctly predict 6 out of 10 teams that made the 2019 playoffs. We believed that we could improve upon our findings by using other machine learning algorithms or including more features that thus increase the overall accuracy of our training model.