imbalanced data

Recently, encountered this question in an interview. Given a data with million rows and 5000 features,how can we reduce the features? It's an imbalanced dataset with 95% positive and 5% negative class (other than using dimensionality reduction techniques)