Mobile Churn Rate

The Challenge

In 2019, an Italian mobile operator expressed its concern about the high churn rate caused by an aggressive pricing strategy of a new entrant MNO in the market. The operator needed to predicting its customer churn for the coming quarters.

As a part of  Machine Learning offering, I was in charge of building a predictive model  which could have helped the operator to predict subscribers churn rate.

The Project

This model provided a machine learning model along with custom Python scripts for solving the churn rate prediction for the operator.

The model focused on binary churn rate prediction, to categories the subscribers as churners or non-churners.

The overall template was divided into 4 major experiments with each containing one of the following steps:

The template model toke two data sets as input: subscribers’ information data set and the activity information data set. Any data following the schema of the subscribers’ information data set and the activity data set can be used with the churn template. Furthermore, this template churn model was generalised to handle different churn definitions (post paid vs. prepaid) on the granularity of time span as input.

The steps to implement this project were as follow:

1- Data Preparation including:
Data Cleaning
Data Labeling

2- Adding New Measures which could have helped in accurate prediction of churn rate, using the existing metrics.

3- Training and Evaluation
The purpose of this step was to train different classifiers and evaluate their performance on the test data.

The Results

In the end Score Model and Evaluate Model modules were used to compare the performance of the Logistic Regression and Boosted Decision Tree.
Our model scored an impressive 0.74 AUC .
Our project report gave the operator a detailed understanding of how various factors such as type of tariffs, packages and age group impact the churn rate and allows the strategy team to plan ahead through our analysis of subscribers’ data.
As the consequence of insightful analysis, the operator saw not only a decline in the churn rate, but also ‘why’.
This is crucial for understanding which the operator’s strategy team to reformulate their marketing strategy.
The ROC Curve, Precision/Recall trends and the Lift Curves for this particular data set can be seen in the following graphs:

0.74

AUC score on the validation set.

Share on twitter
Share on linkedin
Share on email