Team MatrixNotFound · Learn to Hack: MATLAB Edition 2026

Retrofit Radar

We predict how much energy a building should use from its type, size, age and climate, then flag the buildings using far more than similar ones. Those are the first places to audit and retrofit. Built end to end in MATLAB on the WiDS Datathon 2022 data.

▶ Watch the 5-minute demo video

40.1
Leave-one-year-out RMSE (LSBoost v3)
−34%
Error vs. a single regression tree
1,316
Buildings using 50%+ more energy than expected
0
Difference between the browser model and MATLAB
RMSE by held-out year for each model

Results dashboard

Interactive charts of every model iteration, accuracy metrics, the actual-vs-predicted grid, and a filterable retrofit shortlist. It updates live while the pipeline runs.

Open dashboard →
 Predicted energy use of NTU's The Hive and The Arc

3D building simulator

Click NTU's The Hive or The Arc in 3D. Our trained model runs in your browser and shows energy intensity, annual MWh and average kW, with live response curves.

Open simulator →
 Retrofit Radar top 20 buildings

Written report

The full walkthrough, exported from a MATLAB live script: data findings, validation design, model iteration, accuracy, Retrofit Radar and the NTU what-if.

Open report →