Solid work on showing a full Java ML pipeline in action. The feautre engineering approach with win percentages and scoring averages is straightforward, but I dunno if home advantage as a fixed 1.0 is granular enough. Some stadiums have wildly different effects (think Denver altitude or crowd noise variations). I built a similar prediction system last year and found that encoding home field as a team-specific variable improved accuracy by a few percentage points. Still, the clean architecture here makes it easy to iterate on.
Solid work on showing a full Java ML pipeline in action. The feautre engineering approach with win percentages and scoring averages is straightforward, but I dunno if home advantage as a fixed 1.0 is granular enough. Some stadiums have wildly different effects (think Denver altitude or crowd noise variations). I built a similar prediction system last year and found that encoding home field as a team-specific variable improved accuracy by a few percentage points. Still, the clean architecture here makes it easy to iterate on.
Awesome ! Perhaps just a missing line in the pom.xml listing, nothing serious.
What did I miss? Thankfully there's always the full source code on my github repository to take a look :) https://github.com/myfear/ejq_substack_articles/tree/main/nfl-game-predictor