Please use the environment specified in codebase
I think a good place to start is with the data scraped from Kaggle, https://www.kaggle.com/andrewsundberg/college-basketball-dataset/data
I would just plop this in a folder march-madness/data/kaggle-data/ for now
In order to get the data for the 2020 season so far, navigate to codebase/data-scrape/
and then run python Scrape2020.py upon doing so the file cbb20.csv will appear in data/kaggle-data/
I am still looking for more comprehensive datasets, might have to shell out some money for that I might need to find a vegas dataset, look at strategy #3 for why
My hypothesis is that there are really three approaches here:
- Play the bracket:
- Basically break down what matchups in the past have been most successful i.e. how often has a 5 seed beaten a 12 seed in the first round
- Play the teams:
- Based on the season performance of a team, how likely are they to beat their opponent i.e. based on the datset Georgetown is most likely to beat Syracuse becaue of, well math, im not there yet
- Play Vegas:
- For this I would need another dataset, to see how money is moving in vegas
- Basically if the line moves one way that would show that the public and vegas are prepareing for a particular outcome i.e. if the line moved in the Georgetown v. Syracuse game against Georgetown's favor, then it is more likely Georgetown will win
My hope is that this will end up being an ensamble model, as all rounds after the first are inherently unpredictable so to fill out a good bracket, the first round would be based on teams and vegas, then the next rounds would have to rely somewhat on the bracket Just a hypothesis, this is all subject to change.