Data preprocess

• make the columns name in the same format (lowercase + replace ‘ ‘ with _)
• make categorical data in the same format (lowercase + replace ‘ ‘ with _)
• to_numeric → change to numerical format
  • errors → coerce (fix invalid parse as Nan)
• change target variable to numerical (if the original data is in categorical form)

EDA

• view target variable distribution
• view churn rate
• view Nan column

Feature importance

churn rate

• mean of churn
  • churn = 1 → person who churn
  • mean of churn = person who churn / all samples = churn rate

Risk ration (group churn rate / global churn rate) -> relative term

• churn rate of focus feature / global churn rate
• >1 → more likely to churn
• <1 → less likely to churn