Lineup Forecasting

This project is an attempt to determine optimal lineup combinations in the NBA using predictive modeling.

The Problem

NBA lineups are usually easy to understand from a production standpoint if there is on-court data available for a specific player combination. It becomes much harder to predict future performance when there is no historical precedent for the lineup in question. This is a problem given the how common new player combinations are created through the draft and player transactions.

The Solution

This tool creates model which takes in a lineup and returns an associated performance prediction in the form of a winning percentage. This can be used to forecast new and unique player combinations and can even be extended to team performance over a season. The model is fit using player data scraped from Basketball Reference.

In this repository is the following:

• data, directory of the scraped player data
• reports, directory which includes a sweave report detailing the model and its implications
• boosting.R, file fitting the final model
• bref_scraping.R, file for scraping the necessary data

The rest of the repository includes files dealing with various data transformations.