Advancing NFL win prediction: from Pythagorean formulas to machine learning algorithms.

Area of Science:

• Sports Analytics
• Machine Learning
• Predictive Modeling

Background:

• Traditional sports analytics often relies on established formulas like Pythagorean expectation.
• Forecasting NFL team performance is complex due to numerous variables.
• Evaluating predictive model accuracy is crucial for strategic decision-making.

Purpose of the Study:

• To compare the predictive performance of traditional and machine learning models for NFL winning percentages.
• To assess the efficacy of Random Forest regression and Neural Networks against the Pythagorean expectation formula.
• To identify key performance indicators influencing team success through feature importance analysis.

Main Methods:

• Utilized a 21-season NFL dataset (2003-2023).
• Implemented and compared Pythagorean expectation, Random Forest regression, and a feedforward Neural Network.
• Employed performance metrics like Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared.
• Conducted feature importance analysis using SHAP values.

Main Results:

• Machine learning models demonstrated superior predictive accuracy over the Pythagorean method.
• The Neural Network model achieved the highest performance (MAE=0.052, RMSE=0.064, R²=0.891).
• Points scored, points allowed, margin of victory, turnovers, and offensive efficiency were key predictors.

Conclusions:

• Machine learning models offer greater flexibility and robustness than fixed-formula approaches for NFL forecasting.
• Advanced data-driven models enhance decision-making for sports analysts, coaches, and management.
• The findings support the integration of machine learning for optimizing strategic decisions in professional football.