Required packages: tensorflow, jupyterlab, pandas
ANN to select which of two NBA players is better, based on selected data from basketballreference.com. Trained using data from the 2020-2021 NBA season.
Methodology:
- Import every player's per 100 possessions stats for the 2020-2021 season from basketballreference.com, while trimming derived statistics such as shooting percentages, which are already accounted for in shooting attempts and makes.
- Manually label players as either good (1) or bad (0). I was as vague as possible when considering whether a player was "good", to avoid bias. For example I tried to include players that were considered good defensively but bad offensively, like Gobert and Covington, while also having the opposite such as Harden and Trae Young.
- Create matchups by pairing good players with bad players, then randomize them so that the matchups aren't always good player on right, bad player on left.
- Separate the data into test and training sets, and also separate the labels from the actual data.
- Train the neural network with this data.
Note: This project is a work in progress. There are still many things I would like to do with this, such as incorporate Offensive and Defensive rating, test if the network can compare accross eras, etc.