ADS Capstone Chronicles Revised

3

3.1 A predictive analytics model for forecasting outcomes in National Football League games using decision tree and logistic regression. The study explores the broader application of predictive analytics in sports, with a particular focus on predicting NFL game outcomes. Decision trees and logistic regression, common techniques in sports analytics, were employed for predictive modeling. Emphasis was placed on variable importance, particularly in identifying offensive turnovers as a significant predictor of game outcomes. The study acknowledges the potential variability in predictive accuracies resulting from different modeling approaches. Although both decision trees and logistic regression were effective, their comparison revealed differences in performance for NFL game outcome prediction. In their validation, the study found logistic regression outperformed decision trees in predicting NFL game outcomes, highlighting the importance of selecting appropriate modeling techniques for accurate predictions. Despite the comprehensive analysis, the study had limited variable selections. Further performance improvement could be achieved by incorporating additional factors into the analysis (Gifford & Bayrak, 2023). 3.2 Predicting Plays in the National Football League. In contrast to the previous article, this study developed a comprehensive list of variables derived from given variables, ensuring only those conducive to predictive analytics techniques aimed at forecasting plays in NFL games were retained (Joash Fernandes et al., 2020). These variables were carefully selected to capture relevant features from game data, forming the basis for employing various machine learning algorithms for prediction. A notable pattern observed in the study is the utilization of a diverse range of machine learning algorithms, including

classification trees, k-nearest neighbors, random forests, and neural networks, to predict NFL plays. This diversity highlights the flexibility in modeling techniques employed to tackle the predictive task. Furthermore, the study adopted a rigorous approach to testing and validation, employing 10-fold cross-validation over 15 iterations. This iterative process contributes to the reliability and robustness of the study’s findings by thoroughly assessing the performance of predictive models. However, a contrasting aspect arises in the trade-off between prediction accuracy and interpretability observed across different machine learning algorithms. Though decision tree models demonstrated high accuracy, neural networks exhibited superior prediction accuracy at the expense of reduced interpretability. Regarding validation, the study provides insights into the performance of predictive models in forecasting NFL plays. This validation is evidenced by the reported accuracy rates of the decision tree models, validating the effectiveness of the modeling approach. Moreover, the study highlights the practical relevance of the neural network refined model, particularly in time sensitive situations, such as assisting coaches in strategy modification. Despite these strengths, one notable gap in the study is the limited discussion on methods to enhance the interpretability of complex machine learning models, such as neural networks (Joash Fernandes et al., 2020). 3.3 Sports analytics in the NFL: classifying the winner of the Super Bowl. Roumani’s (2022) approach pioneers the incorporation of class imbalance analysis to predict the Super Bowl winner. In the NFL, where only one team emerges victorious each year, the inherent data imbalance often goes unaddressed. Roumani (2022) addresses this gap by introducing the Synthetic Minority Oversampling Technique (SMOTE) to rectify class imbalances in NFL data analysis, specifically for classifying Super Bowl winners. Through experimentation, Roumani

30