A residual is the portion of a measurement left unexplained by a model: approximately the measurement minus its estimate. Modeling residuals focuses the next prediction step on what the baseline misses. section 6.2.1
Subtracting a mean baseline
For measurements 1.52 and 1.21, the mean estimate is 1.365. Subtracting it produces residuals of +0.155 and −0.155. Gaussian process regression then applies kernel-derived weights to these residuals and adds the mean back to form estimates. section 6.2.1
Residual-based estimation follows this pattern:
- Subtract the baseline so the model works on unexplained variation. section 6.2.1
