Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/shrianshchari/nba-predict


https://github.com/shrianshchari/nba-predict

Last synced: about 1 month ago
JSON representation

Awesome Lists containing this project

README 

        
# nba-predict

Here's how I grabbed NBA team statistics off the Internet and condensed them into CSV files for easy analysis.



#### Step 0: My goal

After reading articles about using machine learning to predict the winner of the NBA Finals, I wondered if I could create a predictive model using the four factors of basketball:

- Shooting

- Free throws

- Turnovers

- Rebounds



These are the four main ways possessions end in basketball, and I wanted to know if these statistics in particular could predict an NBA champion.



#### Step 1: Getting my data

I looked around online, and the data I found either didn't fit what I wanted, or was too much for me to use. I just required the four factors stats of each team in the regular season, and I realized that the best place to find that was on [nba.com](https://www.nba.com/stats/teams/four-factors/?sort=W_PCT&dir=-1&Season=2021-22&SeasonType=Regular%20Season). Of course, they didn't provide clean CSV files for me to use, so I had to make them myself.



After doing some testing with BeautifulSoup, here's what I ultimately did:

- Since using BeautifulSoup on the NBA website's URL would just yield the website before rendering, and I wanted the rendered table, I saved all the webpages in my browser using Ctrl + S.

- Then, I used BeautifulSoup to process these saved, rendered webpages and extract the table data into a second HTML document that would only contain the tables.

- Next, I parsed the data within the tables into a pandas DataFrame.

- Finally, I cleaned the data up by removing commas that the NBA includes in their four-digit numbers and I had a perfect dataset, ready for any analysis that I wanted to do.



#### Step 2: Playing with my data

Now that I had my data, I could explore it a bit and see what it looked like. I picked the statistics for the latest season and charted them in order from least to greatest. Then, I wanted to visualize all four graphs at once, so even though none of them were normal curves, I took their z-score and plotted them all on the same graph. All of these plots can be found within the Python notebook within the repo.



#### Step 3: Getting playoffs data

I replicated what I did in step 1, but instead of grabbing regular season data, I would grab playoff data. I didn't feel especially inspired to do any sort of data analysis for this playoff data, especially since most of the teams didn't play more than 5 or 6 games.



I also realize in hindsight that it would've been a lot easier to rip the data from Basketball Reference, or even just download it from Kaggle instead of ripping it from the NBA's website, but I'm also fine with what I did because it gave me good practice with using BeautifulSoup.



#### Step 4: Redo step 1 (for some reason)

I got it in my head that it would be a good idea to just rip my data from Basketball Reference in the future. This made so that if someone else wanted to replicate this for themselves, they could just run the code in the notebook instead of having to go to NBA.com and start saving webpages. After a bit of work, I did exactly that, with a similar procedure to step 1.



The advantage to this is that I get to use more data, since the NBA only had Four Factors data going back to the 1997-1998 season, whereas Basketball Reference has Four Factors data going back to the 1983-1984 season. This gives me access to significantly more data, and will hopefully make my data more accurate.



Also, with the NBA Finals recently concluding, it was a good opportunity to jump back onto this project and see how well this model would work.