Scribd
Upload
556 views

Slide 1: No-Churn Telecom

The document provides information about a project to build machine learning models to help a telecom company called No-Churn reduce its churn rate. The goals are to understand what variables influence customer migration, create churn risk scores to target retention campaigns, and build a model to predict churn. The dataset has over 4,600 rows and 21 columns on customer attributes and usage. Model building steps include EDA, feature selection, model training and validation. The objective is to accurately predict churn and non-churn with high recall to prioritize customer retention. Insights from EDA found account length, voice plans, area codes and states with highest churn.

Uploaded by

leongladxton
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
556 views

Slide 1: No-Churn Telecom

The document provides information about a project to build machine learning models to help a telecom company called No-Churn reduce its churn rate. The goals are to understand what variables influence customer migration, create churn risk scores to target retention campaigns, and build a model to predict churn. The dataset has over 4,600 rows and 21 columns on customer attributes and usage. Model building steps include EDA, feature selection, model training and validation. The objective is to accurately predict churn and non-churn with high recall to prioritize customer retention. Insights from EDA found account length, voice plans, area codes and states with highest churn.

Uploaded by

leongladxton
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 11

Slide 1

No-Churn Telecom
Churn Rate ML Project

Group 123
- Sagar R
- Jayanth Reddy
Datamites - Syed Zuhib Ahmed
Business Case
• No-Churn Telecom is an establish Telecom operator in Europe with
more than a decade in business. Due to new players in the market,
telecom industry has become very competitive and retaining
customers becoming a challenge.
• In spite of No-Churn initiative for reducing tariffs and promoting
more offers, the churn rate (percentage of customers migrating to
competitors) is well above 10% annually.
• No-Churn wants to explore possibility of Machine Learning to help
with following use cases to retain competitive edge in the industry.
Project Goal
Help No-Churn with their use cases with ML
1. Understanding the variables that are influencing the customers to migrate.
2. Creating Churn risk scores that can be indicative to drive retention campaigns.
3. Introduce new predicting variable “CHURN-FLAG” with values YES(1) or NO(0)
so that email campaigns with lucrative offers can be targeted to Churn YES
customers.
4. Exporting the trained model with prediction capability for CHURN-FLAG which
highlights the flag (with input variable documents) that can be integrated with
internal application and which helps to identify possible CHURN-FLAG YES
customers and provide more attention in customer touch point areas, including
customer care support, request fulfilment, auto categorizing tickets as high
priority for quick resolutions and with any questions they may have.
Dataset Description
1. Source dataset is SQL Server in client c_AreaCode q_EveCalls
site. c_InternationalPlan q_EveCharge
2. Dataset contains 4617 rows and 21 c_Phone q_EveMins
columns c_State q_InternationalCharge
3. There is no missing values for the c_VMailPlan q_InternationalMins
provided input dataset. q_AccountLength q_Internationalcalls
q_CustServCalls q_NightCalls
4. y_Churn is the variable which notifies
q_DayCalls q_NightCharge
whether a particular customer is churned
or not. And we will be developing our q_DayCharge q_NightMins
models to predict this variable only. q_DayMins q_VMailMessage
y_Churn
Model Building Steps

Define Problem Data Collection Data Preparation Perform EDA

Validate &
Improve Models
Feature Selection Build Models Measure Models
Performances
Performance

Select the Best Fit


Execute Models
Model for our
for Prediction
Business Case
Model Objective
1. Predict Churn and No Churn
2. Decision Function is higher Recall rate
3. Performance Function is higher Accuracy Score
Exploratory Data Analysis - Overview
Slide 2
Churn and No Churn International Plan Area Code vs Churn

14% 10%

Churn Yes
NoChurn No

86% 90%

Customer Service Calls vs Churn (%)


State vs Churn
Slide 2

WV
3.227

2
Business Insights
1. Account length of duration 60 to 140 days has higher churn.
Highest churn is in the range of 100-120 days.
2. People without voice message plan are 86% and with the plan are
14%. We can imply that if we offer more voice message plan it
can help us reduce the churn.
3. 42% of Churn are from Area Code 415 and the top 5 churn are
from WV, AL, MN, ID and OH states.
4. After 3 customer service calls, the churn rate spikes up.
5. Cut-off for churn rate has been set to 0.18% which gives the
highest Recall of 88% and efficiency of 95%.
Modelling with Python
Slide 11

THANK YOU

You might also like