Telecom-Churn

• This project involves building a binary classification model for predicting telecom churn, a well-known and fundamental classification problem.

• The notebook includes an overall machine learning process, including Data cleaning and visualization, new features and feature selection, baseline model, model selection and evaluation, feature importance, hyperparameter tuning, model interpretation and final model.

• Specifically, the major tools included:

  • violin plot to check the numerical data vs label
  • countplot to check the catagorical data vs label
  • check features correlation
  • split train and test data
  • apply onehotencoder and standardscaler to fit and transform on train set
  • apply smote to address the imbalanced data
  • define baseline model
  • fit and predict model
  • output classification report
  • check the confusion matrix
  • build in pipeline
  • Test in different models
  • find the best for corss validation and hyperparameter tuning
  • plot ROC curves
  • check the feature importance using permutation importance
  • Run the final model and predict on the holdout set