Disclaimer: This project is not an endorsement of gambling, and the insights provided are not meant to be taken as betting advice.
In recent years, sports betting has exploded in popularity. However, rarely are the odds offered on a particular wager a true reflection of the probability of that event happening, but rather they are an attempt by bookkeepers to 'balance the action' on both sides of a wager such that a profit is made regardless of outcome. Bettors therefore need alternative means of predicting a given proposition's outcome beyond the consensus odds being offered. In this project, we will explore the viability of predicting the results of points-based proposition wagers in the NBA. A proposition wager is a wager on an event that is independent of the game's result (e.g. Player X will score above or below 30 points). Using a combination of player and opposing team statistics, as well as Vegas betting metrics for a given proposition, we will attempt predict player performance, use those predictions to inform a betting strategy, and evaluate both the accuracy of the model, and whether it would have won or lost money.
The goal of this project is to assess the viability of using linear regression to predict player scoring performance in an NBA game, and how those predictions fare against publicly available betting lines from the past, and real-world wagers for upcoming NBA games.
We will use two datasets to gather our features and target variables:
- Betting information taken from BettingPros.com
- Player/Team Statistics from Stathead.com
- Selenium and Beautiful Soup for Web Scraping
- Pandas and NumPy for Data Ingestion, EDA
- Seaborn for Visualization
- Scikit-learn for regession analysis and model testing.
Produce a baseline model using all of the features available, as well as testing metrics to inform further iterations.