This project is no longer under development. See my more recent nba repo
for my current efforts on this subject.
This project (beginning around the time of my graduation from college in May 2016) was my first real dive into experimenting with machine learning models. I initiated this project as a means of re-inforcing the concept that I had learned in the Introduction to Statistical Learning with R (ISLR) book. The proejct itself explores the concept of improving NBA team win total predictions using team-level statistics, Las Vegas odds, etc. After developing some scripts to scrape data from websites like (https://www.basketball-reference.com), I tried out various classification and linear regression techniques to make predictions.
The primary deliverables of this projects are the classification.R and linear_regression.R scripts in the R folder. Both are formatted such that they can be converted directly to a .html or .pdf file
(using the rmarkdown::render() function). This was done with classification.R. (See the
output folder). On the other hand, the linear_regression.R script was never fully completed.
While there was no real goal of discovering a method to improve NBA win total predictions, I found that no single classification method stood out as the best. In regards to predictors, it appears that the error in the Las Vegas total for the previous season and the implied odds (derived from the Las Vegas total) for a given season serve as the best predictors.
This project was heavily restuctured when cleaning it up for GitHub. (Actually two projects were combined into one to create this one.) The code that is in this repo cannot be expected to run properly without some revising.
Also, this project was my first ever project with R programming, so the coding style is probably something that should be emulated.