I've developed a model that performs fantastic in back testing but, for lack of a better term, sucks in live testing. I drilled down the issue to the fact that I only fed the model the players who actually played in the game versus the team's full roster. For example, the Brooklyn Nets have 17 players on the roster but last night only 10 players actually played in the game. Is there a decent source for this information? Rotowire has "expected lineups" but afaik they only include starters. I've been thinking about creating another model that predicts if a given player will play in a given game or not.
I've developed a model that performs fantastic in back testing but, for lack of a better term, sucks in live testing. I drilled down the issue to the fact that I only fed the model the players who actually played in the game versus the team's full roster. For example, the Brooklyn Nets have 17 players on the roster but last night only 10 players actually played in the game. Is there a decent source for this information? Rotowire has "expected lineups" but afaik they only include starters. I've been thinking about creating another model that predicts if a given player will play in a given game or not.
I was able to solve the issue by creating a model that predicted if a player was going to play in a game or not. At first the accuracy was 82% but then I was able to get it to 95% by scraping the nba injury/suspension data from the prosportstransactions.com database. The model is back to performing fantastically!
I was able to solve the issue by creating a model that predicted if a player was going to play in a game or not. At first the accuracy was 82% but then I was able to get it to 95% by scraping the nba injury/suspension data from the prosportstransactions.com database. The model is back to performing fantastically!