Basketball Data Source: https://www.basketball-reference.com
Year Timeline: 2015-16 till 2023-24 NBA Season (9 seasons)
For ML model training, it is best to use as many data points and datasets (season years) as possible for optimal accuracy.
However, given the nature of the game, I had to also consider the relevance of the timeframe I was choosing (ex. a game from 20 years ago would not be relevant to predict a game from this year).
Using both technical research and my knowledge of the game, I decided to keep the 2015 - 2016 NBA Season as the starting point of my web scraping as it marks several key trends and shifts in the NBA that are relevant to modern basketball, including:
- Three-Point Revolution: The emphasis on three-point shooting and "spreading the floor out" became much more pronounced around this time, largely influenced by the success of Stephen Curry and the Golden State Warriors in the 2014-15 season.
- Advanced Metrics: Increased reliance on advanced metrics and analytics to guide team strategies and player evaluations.
- Player Mobility: Greater player movement through free agency and trades, leading to the formation of super teams and significant shifts in team dynamics.
- Rule Changes: Various rule changes aimed at increasing the pace of the game and scoring.
As of now, my base model has an accuracy of 63% using data from over 23,000+ games. My next steps to increase accuracy are: (1) use xgboost and/or random forest classifier, which is generally more powerful than ridge regression, (2) experiment with the different feature numbers, (3) utilize the backwards feature selector rather than forward.
I would like to credit Dataquest on YouTube for ML and web-scraping related lessons, which helped me gain a better understanding of developing the model.