Feature Selection for Classification Using Deep Reinforcement Learning

A review of feature selection methods: traditional methods v.s. DRL methods

Image by sloppyperfectionist on Unsplash

I am opening up a review series to help scholars/students collect research ideas.

In this series, I will discuss a research topic and then state the motivations, related works, methods, and experimental results by reviewing relevant papers.

Today’s topic is Feature Selection for Classification Using Deep Reinforcement Learning.

In this blog, you will learn:

• Traditional algorithms features selection: filter methods (e.g., univariate feature selection, correlation-based feature selection); wrapper methods (e.g., evolutionary algorithms, branch and bound algorithms, greedy search algorithms); embedded methods (e.g., LASSO, decision tree).
• Motivations to use DRL algorithms to perform feature selection.
• MDP settings for feature selection
• Open-sourced packages to implement PPO for feature selection

1. Motivations: