Fig. 4: The predictions of slavery prevalence (individuals enslaved as a % of the population) made by the best model and the five other well performing models in the Rashomon set (see Table 2).
From: Machine learning methods for “wicked” problems: exploring the complex drivers of modern slavery
The x-axis is ordered by the MAE between our best model and the “actual” prevalence as estimated by the GWP survey data.
