Telecom Churn Prediction

• General Info
• Technologies Used
• Conclusions
• Acknowledgements

Overview Telecome is a very competitive market, exit barriers for customer are low. Loyalty depends on user experince and attractive offers. Industry experiences an average of 15-25% annual churn rate. Given the fact that it costs 5-10 times more to acquire a new customer than to retain an existing one, customer retention has now become even more important than customer acquisition. Retaining high profitable customers is the number one business goal.

Problem Statement To reduce customer churn, telecom companies need to predict which customers are at high risk of churn.

• The goal is to build a predictive model that can predict the chrun propensity for a customer accurately.
• Identify key features that indicates the customer behaviour to churn

Approach

• Data understanding from data dictionary
• Loading the selecting the relevent data for analysis
• Data sanity check - check for duplicates, cardinality, format mismatch and missing values
• Data cleaning - Use of appropriate strategy for imputing missing values and extreme/outlier values
• EDA - Univariate, Segmented and Bi-variate analysis with Visualization
• Feature Engineering - Using the termporal behavioural data creating new features to capture the change in behaviours like Rate of Change and Direction of Change, extrating features from date-time features
• Data Preparation - Scaling of data
• Model development experiments - 2 modeling approaches

  • Model development to understand the important features wrt target variable

    • Obtain Top K Feature Importance using Random Forest Mode
    • Use Logistic Regression to develop explinability using Feature Importance

  • Model development for accurate prediction

    • PCA is used for dimentionality reduction
    • Various algorithms will be fitted - Logistic Regression, Random Forest, XGBoost, LightGBM and Ensemble with Hyper-paramter Tuning
• Conclusion

• With good feature engineering of the temporal data the model prediction can be further imporved

• Using Principal Componets for prediction has not resulted in higher prediction performance of the model even after hyper-parameter tuning compared to using raw features as is.

• As for the explainablity concerned;

  • There is a clear pattern across last 3 months (month 6, 7, and 8) with respect to the call usage, data usage and recharge behaviour of the customer with higher probaility to churn
  • The customer with higher probability to churn exhibit following behaviour as per the data;
    • have reduced recharge amount and total recharge from month 6 to month 8
    • The Average Revenue Per User has significant drop from month 6 to 8
    • The customer with higher median Roaming calls (both incoming and outgoing) MOU has higher tendency to churn
    • Total outgoing call MOU also has significant drop from month 6 to 8
    • Customer who are with network for longer than 1000 days has low tendancy to churn

• These features can be a good indicator for probability to churn

• Recommendations

  • Offering discounts on Roaming Charges or special Roaming packages may prevent customer have high roaming usage
  • Offering special discounts on Month 7 and 8 on every recharge could help retaining the customer
  • Offering special packages to customer who are less than 1000 days old

• python - 3.10.1
• pandas - 1.4.2
• numpy - 1.21.5
• matplotlib - 3.5.1
• seaborn - 0.11.2
• scikit-learn - 1.0.2
• scipy - 1.7.3
• statmodels - 0.13.1
• imbalanced-learn - 0.9.0
• xgboost - 2.0
• lightgbm - 3.3.1

Give credit here.

• This project was inspired by...
• References if any...
• This project was based on this tutorial.

Created by [@githubusername] - feel free to contact me!