Predictive Analytics
Why there’s no “trained model” here yet
The founding research covers six countries. Six data points cannot support a train/test split, and a Random Forest or gradient-boosted model fit on all six would simply memorize them — it would report high accuracy while generalizing to nothing. Building that and calling it “predictive” would look impressive and be false.
What’s actually defensible at this sample size: simple single-variable linear regression, reported honestly as a descriptive measure of association within these six countries — not a validated, generalizable model. As this platform adds more countries to the dataset, this page will graduate to genuine multivariate regression with holdout validation.
Which single factor tracks child labor most closely?
R² from a simple linear regression of each indicator against the actual child labor rate, across the six studied countries.
Education spending (% of GDP) shows the strongest single-factor relationship (R² = 0.41, r = -0.64) — consistent with the founding paper’s own conclusion that public investment in education, not just household income, is what actually moves child labor outcomes. Urbanization shows almost no relationship on its own, which lines up with the Risk Index finding that India and Bangladesh — both still mostly rural — keep child labor low through policy, not urban development.