RESIDUAL ANALYSIS
A deeper look at where and how the model makes errors
What is a residual?
A residual is simply the difference between what the model predicted and what the actual value was. For example, if the model predicted 1.20 but the real value was 1.15, the residual is +0.05. Studying residuals tells us whether the model makes random small errors (good) or systematic biased errors (bad). A good model has residuals scattered randomly around zero with no pattern.
Error histogram — are mistakes small and symmetric?
This bar chart shows how often the model made errors of different sizes. Green bars = over-predicted, red bars = under-predicted. Ideally the chart looks like a bell curve centred on zero — meaning most errors are tiny and there's no consistent bias in either direction.
Residual Distribution · Selected Property
ERROR HISTOGRAM
Residuals vs fitted — does the model struggle at extremes?
Each dot is one sample. The horizontal axis shows the model's predicted value; the vertical axis shows the error. Dots should scatter randomly around the zero line with no funnel or curve shape. A funnel shape would mean the model gets worse at extreme values.
Residuals vs Fitted Values
RESIDUAL PLOT
Mean Absolute Error — all properties
MAE PER PROPERTY