Comparative Machine Learning Analysis Highlights Novel Predictive Capability of Deep Neural Decision Forest for Ulnar Collateral Ligament Reconstruction in Baseball Athletes

Ulnar collateral ligament (UCL) reconstruction, commonly known as Tommy John surgery, has become increasingly prevalent among Major League Baseball (MLB) pitchers. We hypothesized that advanced machine learning (ML) techniques combining deep learning with ensemble decision trees could accurately predict the likelihood of UCL reconstruction using performance metrics from public databases. To test this, we employed a matched-pair design, selecting MLB pitchers with a minimum of 1500 career pitches and compiling diverse player profiles from Baseball Savant and Baseball Reference. We developed and compared nine ML models, including logistic regression, K-nearest neighbors, support vector machine, decision tree, random forest, XGBoost, artificial neural network (ANN), Deep Neural Decision Tree (DNDT), and Deep Neural Decision Forest (DNDF). Our findings indicate that while traditional models like logistic regression and K-nearest neighbors achieved accuracies around 62.5% and random forest reached 71.9%, the DNDF model significantly outperformed others with an accuracy of 79.2%. Feature importance analysis, using Shapley values, revealed that total pitch count was the most critical predictor, emphasizing its biomechanical impact on elbow stress. These results suggest that the DNDF model might effectively identify predictive patterns in low-dimensional tabular data, outperforming other ML approaches in UCL injury risk prediction. Future research may focus on integrating additional performance metrics, including pre-MLB pitch counts and real-time sensor data, to enhance prediction accuracy further and promote equitable healthcare interventions for athletes.