Feature Crossing



  • Feature crossing is one of the feature engineering technique.
  • Feature crossing can capture the interaction between two or more categorical features.
  • Feature crossing helps model train faster, easier to predict target label.
  • Feature crossing can degrade learning performance.
  • Feature crossing can cause sparsity.

Pros and Cons

  • Help model to make prediction easier
    • For example, the cross feature ‘job ⊗ company’ indicates that an individual takes a specific job in a particular company and is a strong feature to predict one’s income
  • Faster training time → Reduce training cost
  • Cross features are highly interpretable
  • Can degrade learning performance
    • Irrelevant or redundant or introduce noise
  • Cause sparsity

  • Handling numerical features
    • Bucketize or convert into the categorical feature before creating a feature cross
  • Handling high cardinality
    • Embedded them into an embedding layer to reduce the dimension

Additional Resources


#machine-learning #feature-cross