Steam Games Performance Visualizer & MLP Regressor
Nov 2025
An end-to-end study of the Steam catalogue: clean a hundred thousand rows, work out what actually predicts a game's performance, then let someone put a question to the model in plain terms and get an answer back with the chart that supports it.
- Processed and modelled a 100k-row dataset spanning 1,260+ unique games in pandas and Matplotlib — normalisation, categorical encoding, and outlier handling included.
- Designed interactive, accessible dashboards surfacing the performance signals that matter for game analytics, so evaluation rested on evidence rather than instinct.
- Developed branching logic driven by qualitative metric analysis to train targeted scikit-learn MLP regression models against user-defined queries, holding R² above 0.50 across every model.
- Language
- Python
- Modelling
- scikit-learn
- Data
- pandas
- Visualisation
- Matplotlib · Seaborn
I rebuilt the modelling side of this as something you can actually run. The sandbox trains a multilayer perceptron in your browser — written from scratch in JavaScript, no libraries — and races it against a linear regression on the same data, so you can see where the extra machinery earns its place and where it does not. Held-out R², a live loss curve, and permutation feature importances are all computed as it trains.
Open the live sandbox Train the network yourself — runs entirely in the browser