Predictors of Increased Poverty
The model will set default weights using the Random Forest Classifier
feature importance based on an increase of 2% or more for the selected indicator.
Random Forest Classifier
Values are per 1,000 individuals.
Disclaimer: Data is based on projections which is prone to error.
2016 predicting 2017 Poverty
For model training, a "y" column value of 1 indicated zip codes where poverty increased in 2017 after increasing or remaining steady from 2015 to 2016.
Best Params:
max depth: 8;
n-estimators: 100
Accuracy before tuning: 69%.
Accuracy after tuning: 71%.