Fig. 6: Calibration analysis for the hotspot scores calculated on the Beijing dataset.
From: Detecting urban PM2.5 hotspots with mobile sensing and Gaussian process regression
Reliability diagrams (a, c) compare the predicted hotspot scores against the observed exceedance proportions, before (in blue) and after (in pink) an isotonic regression calibration is applied. Isotonic regression fits (b, d) illustrate how the isotonic regression fit transforms the hotspot scores (x-axis) into calibrated hotspot scores (y-axis).
