Let's say the data has 10 features (columns), so, I need to create/make/calculate a new column as a score or ranks based on these features not using traditional way like z-score or normalised or scaled. Just I am interested that how or what machine learning algorithm (unsupervised learning algorithm) can do this? Thank you
1 Answers
One example of deriving unsupervised feature importance is using PCA. PCA would reduce the dataset to 9 features (aka Principal components). The impact that each original feature has on the principal components is easily extracted using any of the available packages e.g., Statsmodels or Sklearn. Both which provide good documentation and examples on how to do this.
Related