Customer Churn Prediction & Analysis for Telecom Data.

Description

The objective of the project is to build a model to predict the probability of the customer churning from the platfrom for a telecom data.
Dataset Source : This dataset is based on IBM based sample dataset obtained from Kaggle for Customer Churn.
Dimensions for the DataSet [rows vs columns]: 7043 * 21.
The Data set includes information about:

• Customers who left within the last month – the column is called Churn.
  
• Services that each customer has signed up for – phone, multiple lines, internet, online security, online backup, device protection, tech support, and streaming TV and movies.
  
• Customer account information – how long they’ve been a customer, contract, payment method, paperless billing, monthly charges, and total charges.
  
• Demographic info about customers – gender, age range, and if they have partners and dependents.
  

Environment

• R Studio, R (ggplot2,dplyr,randomForest,corrplot)

Tasks performed:

• Data Cleaning
• EDA
• Feature Engineering and Optimization
• Data Modeling.
• We implemented Logistic Regresion and found an accuracy of 79.01%.
• Further we tried implementing the Decision Tree classifier but did not find any improvement in our accuracy.
• Lastly we performed random forest claasifier and we observed an accuracy of 80.07%

Findings

Output:

1. The data includes almost equal proportion of males and females.
2. Almost 58% customers are on paperless billing.
3. 26% of the customers have churned from the platform.

Output:

1. Almost 40% of the customers have subscribed for the Fibre optic internet service.
2. Almost 50% of the customers have no online security and almost 45% customers have no online backup.
3. Almost 50% customers have no techsupport access and 40% have no streamingtv as a service.
4. 45% of the customers have no service of device protection.

Output:

1. Maximum number of customers have subscribered for electronic check for their payments.
2. Very less i.e approx 20% of the customers are senior citizens.
3. Equal number of customers with and without partners.
4. 65% of the customers have no dependents.
5. Almost 87% of the customers are with the phoneservice.

Output:

1. Maximum Customers churned from the platform are the one having a tenure of 0-1 years.
2. Maximum Churned customers have a Monthly charge more than $65.

Output:

* Churn rate is equally divided among the male and female customers.
* Churn rate is more among the customers with no dependents.
* Churn rate is more with customers having phone service.
* Churn rate is more with customers having paperless billing.
* Churn rate is more with customers having electronic check as the payment mode.

Output:

* Churn rate is more with customers having month to month contract.
* Churn rate is more with customers having no online security and techsupport.
* Churn rate is almost equal among the subscribers with or without the streamingtv.

Output:

With the above Logistic regression kmodel we could see an accuracy of 78% in the model.

Output:

**2.DECISION TREE**

Output:

3. Random Forest Modeling

Output:

Accuracy Comparison for the three Models

Accuracy for Logistic Model 0.7901063
Accuracy for Decision Tree Model 0.7928803
Accuracy for Random Forest Model 0.8007397

ROC analysis for the three Models

Output:

BUSINESS COST ASSUMPTION :

Output:

Results

• We successfully implemented three classification models in order to predict the potential churn customers. Considering the compares the results aforementioned business cost assumptions Random Forest Model can be considered to be a best fit model of the three implemented models with the best accuracy.
  
• Based upon the descriptive analysis , a greater number of Churn customers are observed in case of customers with Tenure in between 0-1 year and with increase in the Monthly Charges there is substantial increase in the Churn Rate.
  
• Customer on Month to Month contract are more susceptible to Churn.
  
• 22% of the total customers were observed with No Internet Service and thus customers are missing the exclusive services of OnlineBackup, StreamingTV, Online Security and Device Protection.