Skip to content
← All work

Machine learning · Nov 2025

Steam Game Popularity

Random Forest models predicting game ownership and positive reviews from Steam data.

Data
339 games from MySQL; 270 modeled after an IQR outlier filter
Models
Random Forest: ownership (R² 0.55) and positive reviews (R² 0.86)
  • Python
  • MySQL
  • scikit-learn

The question

Do deeper discounts make a game more popular, or is popularity driven by something else? I trained two Random Forest models in scikit-learn: one predicting how many people own a game, and one predicting its positive reviews.

What I found

Discounts barely matter. Review counts and concurrent players predict ownership far better than price or discount size, so deeper discounts alone are a weak growth lever.

What predicts how many people own a game

Feature importance from the ownership model (R² 0.55). The reviews model (R² 0.86) tells the same story: concurrent players and owners drive reviews, and discount is last.