Figure 5. Cross-validation scores for (A–J) the top-10 perceived importance ranked prediction tasks (Perceived Importance Ranking) and (K–O) the top-five perceived learnable prediction tasks (Perceived Learnability Ranking).

Performance variability was assessed with bootstrap replicates of the crowdsourced datasets and class imbalance was assessed by randomizing the target variable relative to the input data. At least two of the perceived importance tasks and four of the perceived learnable tasks—highlighted in green—demonstrate significant learnability (predictive performance) over random (Cohen’s d > 2). Note that the regression task in (F) showed poor predictive performance for reasons we describe in the discussion.