Fig. 1

Global prediction accuracy. Boxplots report the distribution of AC prediction accuracy for nine ML approaches across 100 activity classes on the basis of A BA and B MCC values. The models were built based on randomly selected training and test sets (that is, under “data-leakage possibly included” conditions; see Methods section). In a boxplot, a value distribution is represented by its maximum (upper whisker), upper quartile (upper boundary of the box), median (horizontal line), lower quartile (lower boundary of the box) and its minimum (lower whisker). Individual values representing statistical outliers are shown as black dots