Subsets of the data were then created based on our model's most pertinent features that we considered easily identifiable when evaluating families. We then populated a matrix that could report the predicted probability of success given every possible combination of child and family characteristics.