Figure 4

Distributions of observed deviations for two predicted classes of deviations: small vs. large. Leave One Out Cross-Validation (LOOCV) results were separated into two groups (i.e. ‘good’ and ‘bad’), based on a probability cutoff of 0.2. First and second rows used logistic regression models combining features indicated as row labels. Third row uses the ‘max’ approach to combine models used in the top two rows. Overlaid on top of each distribution, lower, middle, and upper line segments represent 25th, median, and 75th quartiles, respectively.