Fig. 1

Prediction performance gains/losses with different combinations of classifiers and feature selection strategies on all experiments. A Accuracy. B ARI. C Macro F1. The performance gains/losses for all combinations are illustrated by the heatmap. The heatmaps are sorted by the average values of the rows and the columns, and thus, the entry at the top left corner represents the most performance gain combination. The boxplots on the right and bottom sides illustrate the marginal performance gains/losses from classifiers and feature selection methods. The red dotted lines in the boxplots are reference lines at 0 (no gain nor loss)