Vertte

Feature selection methods (in Feature Engineering) 본문

Data-Science

Feature selection methods (in Feature Engineering)

vertte epsilon- 2020. 11. 18. 00:38

data science process

  • Project Scoping(Define Problem)

  • Data Collection

  • EDA

  • Data Preprocessing

  • Feature Engineering

  • Modeling

  • Evaluation

  • Project Delivery / Insights

Feature Engineering

  • Feature selection
    • Filtered (using statistical skills)
    • Wrapper (set certain features ,repeat evaluation ,choose better combination)
      • forward 
      • backward
      • etc
    • Embedded (depending on model )
  • Feature Extracction (PCA..)
  • Feature Generation (using domain knowledge) about feature engineering overview

about many method

  • Filtered method
    low variation(near zero)
    high correlation
  1. save all possible combination correlations (if correlation of A & B is high)
  2. select one of two var , see another correrations of two with another variables execpting the mutual (select A )
  3. repeat 1,2

'Data-Science' 카테고리의 다른 글

Feature selection methods (in Feature Engineering)  (0) 2020.11.18
0 Comments
댓글쓰기 폼