Visualizing Uncertainty in Predictive Models - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 64

Fig. 6.3 Predictions made by model about occupation type in the census domain, projected using
PCA and displayed with pie glyphs ( left ) and speckle glyphs ( right ). Legend shown on far right
These factors indicate that this model exhibits higher error and less confidence over-
all than the income model. A few specific patterns become apparent upon closer
inspection: for example, in the lower right quadrant, there is a cluster of instances
for which the transport-moving (royal blue) and craft-repair (pure blue) classes are
conflated, and another cluster of instances for which the prof-specialty (bright green)
and exec-managerial (olive green) classes are conflated. This is intuitively unsurpris-
ing (since those occupations involve similar skills and training), but the visualization
also enables identifying some more surprising errors (e.g., to the far left just above the