Churn Rate Prediction

Introduction

Churn rate is the percentage of subscribers to a service that discontinue their subscription to that service in a given time period. It is a possible indicator of customer dissatisfaction, cheaper and/or better offers from the competition, more successful sales and/or marketing by the competition, or reasons having to do with the customer life cycle. In this project, we will use the data from a telecom company to predict the churn rate of its customers.

Data Description

• The data is from Kaggle: https://www.kaggle.com/datasets/imsparsh/churn-risk-rate-hackerearth-ml
• The data includes 36992 rows, 21 columns, and 1 target variable. The target variable is churn_risk_score, which is a categorical variable with 5 levels. The other 20 columns are features, including 3 numerical variables and 17 categorical variables.
• The data is imbalanced. The churn rate is 4.2%.

Data Preprocessing

Missing Values

• There are 3 numerical variables and 17 categorical variables. The missing values are filled with the mode of the categorical variables and the mean of the numerical variables.

Outliers

• The outliers are removed by using the Robust Scaler.

Imbalanced Data

• The imbalanced data is handled by using the SMOTE method.

Feature Engineering

• The feature engineering includes:
  • Feature extraction
  • Encoding the categorical variables
  • Feature selection
  • Feature scaling

Model Building

Model Selection

• The models used in this project are:
  • SVM
  • Decision Tree
  • Random Forest
  • XGBoost
  • CatBoost
• The model selection is based on the recall score.

Model Evaluation

The evaluation metrics are: - Accuracy - Precision - Recall - F1 score - Confusion Matrix

Model Tuning

• The hyperparameter are tuned by using the GridSearchCV method.