Diligence leverages machine learning models trained on utility-billed data from over 18 billion square feet of real estate to estimate energy use and carbon emissions. These models achieve over 90% R-squared, meaning they accurately capture the relationship between predicted and actual values across the dataset. Estimates also include a confidence label based on average error margin by property type.
- R-squared (R²) measures overall model fit. A value of 0.90+ means predictions closely track real trends.
- Error margin reflects how far off an individual estimate might be—e.g., a “High” confidence score means most estimates fall within ±25% of actual values.
Example: If actual energy use is 100 units, a model with a 20% error margin will typically predict between 80–120 units. A high R² tells us the model follows real-world patterns; the error margin tells us how precise each estimate is.
Data Updates
Diligence models are retrained monthly using newly available data, ensuring the results reflect the most current building performance trends.