Decoding NBA Positions by Shots

Project Overview

This project seeks to classify NBA players into positions based on their shooting patterns using generative probabilistic models. We explore whether the traditional basketball positions are becoming obsolete and how roles have evolved over time across different NBA eras.

Data

We used shot data from the NBA from the 1997-1998 season onwards, focusing on shot distance and other characteristics, while excluding shot quality metrics.

Methodology

• Probabilistic generative models (EDDA & MDA) for position classification.
• Data preprocessing included missing value treatment and variable selection.
• Exploratory data analysis to understand shooting patterns among different positions.
• Cluster analysis to identify natural groupings in the data.
• Model-based classification to predict player positions.

Results

Our models achieved an accuracy of over 70%, indicating a successful classification of NBA players by shooting patterns. The analysis also confirmed an evolution in the roles of players over the years.

How to Use

1. Clone the repository.
2. Navigate to the code directory.
3. Run the code.R script for the analysis.

Contributing

We welcome contributions and suggestions. Please open an issue or submit a pull request for any improvements.

Authors

• Julius Maliwat
• Giacomo Rabuzzi
• Andrea Robbiani

Acknowledgments

Thanks to Basketball Reference for providing the data used in this analysis.