The US Census dataset contains personal (but anonymized) information about 300 000 american inhabitants. The goals of this mini-project is to predict, given an inhabitant's characteristic, whether she/he will earn more or less than $50 000 a year.
We start by removing unnecessary variables, and explore the characteristics of people who earn more than $50 000 a year. We try to avoid keeping variables which are too much correlated. We do another preprocessing part:
- Continuous variables are normalized (remove mean and divide by the standard deviation)
- Categorical variables are "dummified"
We try several classification algorithms
We use first accuracy as a way to check the validity of the predictions. We also use recall and F1-score.