Telecom Churn Prediction

A leading telecom firm from Southeast Asia has collected customer-level data pertaining to the use of its services for a span of four consecutive months encoded as 6, 7, 8 and 9.

The goal of the analytics project is to

• predict customer churn in the 9th month using data from the preceding 3 months, and
• identify the important predictors of customer churn.

The findings will be used to develop strategies to manage customer churn.

Analysis is performed as per the following steps:

Data Cleaning, Feature Engineering & EDA

• Data Understanding & Cleaning
• Feature Engineering
• Exploratory Data Analysis
• Data Preparation

Prediction

• Decision Tree Classifier
• Random Forest Classifier
• XGBoost Classifier
• Model Selection

Inference

• Model Building
• Model Evaluation
• Model Interpretation

Strategies to Manage Churn

The XGBoost classifier is finalised from among the 3 prediction models because it has the best cross-validation ROC-AUC and Recall scores of 0.895 and 0.78.

The multiple logistic regression model finalised for inference has 19 explanatory variables and the following performance metrics: