Automated Feature Engineering Using Neural Networks

How to automate and greatly improve one of the most tedious steps in data modeling.
Machine Learning
Michael Malin
February 26, 2021

Few will deny that feature engineering is one of the most important steps for producing accurate models. Some people love it, but I am definitely not one of them. I find this step very tedious and it is my firm belief that anything tedious can be automated. While my solution doesn’t entirely eliminate the need for manual work, it does greatly reduce it and produces better results. It also produces a model that consistently beats gradient boosted methods on structured datasets.

This article will answer the following questions:

  • What exactly is being automated?
  • How does it work?
  • How do you build it?
  • How does it compare to other models?
Michael Malin
LinkedIn

With over 12 years of experience, I have deployed over 40 successful projects across all AI domains including computer vison, NPL, GNN, and forecasting. I specialize in TensorFlow graph and deep learning networks.