Fig 1. Boxplots, displaying the estimated generalisation performance over 100 test partitions.

Results for the k-NN transfer learning algorithm applied with (i) optimised class-specific weights (combined), (ii) only primary data and (iii) only auxiliary data, for each dataset.