Data Analytics Project

FROM STATISTICS TO SALARIES: MACHINE LEARNING PREDICTIONS IN THE WNBA

Overview

In this project, I used machine learning to explore whether WNBA player salaries align with on-court performance in the 2024 season. With growing attention on equity in women’s sports, I wanted to see how well player metrics like Player Efficiency Rating (PER), Points Per Possession (PPP), and Win Shares (WS) could predict salary, and what that says about fairness in the league.

Tools Used

Language: R
Libraries: glmnet, GNN, ggplot2, caret, dplyr
Models: Linear Regression, Ridge, Lasso, K-nearest Neighbors (KNN)
Evaluation: RMSE (Root Mean Squared Error)

Process

Using publicly available data from Basketball Reference and Her Hoops Stats, I combined salary figures with performance statistics such as Player Efficiency Rating (PER), Points Per Possession (PPP), Win Shares (WS), and Minutes Played (MP), focusing only on players who logged over 1,000 minutes to capture meaningful analysis.

After cleaning and formatting the datasets, I engineered new features including interaction terms like PER × MP and WS × PPP, and created polynomial features to capture potential nonlinearities. I also applied a log transformation to salary values to normalize the distribution.

I then split the dataset into training and test sets, and I implemented four predictive models: Linear Regression, Ridge Regression, Lasso Regression, and K-Nearest Neighbors (KNN). I then evaulated aach model was evaluated using Root Mean Squared Error (RMSE) to compare predictive accuracy on the test data.

Visualizing salary against PPP, PER, and WS helped surface early trends, and revealed which players stood out as under- or over-valued before modeling began.

Results

The results showed that K-Nearest Neighbors outperformed all other models, with the lowest RMSE and the most accurate predictions. It handled the complexity of the data better than the linear and regularized models, which struggled to fully capture the nonlinear relationships between performance and salary.

The KNN model gave the clearest picture of how metrics like PER, WS, and PPP relate to compensation, and also made space for outliers to be seen more clearly. It was especially effective in highlighting cases like Aliyah Boston, whose salary didn’t match her performance, and Diana Taurasi, whose high compensation seemed more tied to legacy than stats.

Reflection

This project gave me a deeper understanding of how player performance translates into salary in professional sports.

While the models provided solid predictions based on on-court metrics, it quickly became clear that salaries are shaped by more than just stats. Marketability, legacy, fan engagement, and team dynamics all play a role in how athletes are valued. This analysis doesn’t capture those layers, but it was a starting point, a way to explore how far performance data alone can take us.

I’ve always loved women’s sports, and this project was a way to bring that personal interest into my work with data! As someone who’s passionate about using data to uncover patterns and tell meaningful stories, this felt like an exciting and honest glimpse into a much, much bigger conversation.

There’s still so much more to understand, but this project helped me see how tools like machine learning can bring clarity to complex, human questions -- and how they might be used to support spaces that deserve more attention.

INDEX NEXT