Churn-project

Customer churn is a SaaS business metric that measures the amount of customers, accounts, contracts, bookings, etc. that a business has lost over a period of time. Also known as the rate of attrition or just plain “churn”, customer churn is one of the most widely-tracked and heavily-discussed subscription company metrics.

In this project, we used the data of a telecommunication company in order to do explorative data analysis and to apply a model that can predict if a customer is going to leave the business or not.

• The data consists of the following variables :

1. age Represents the age of a customer
2. gender Represents the gender of a customer
3. security_no Represents a unique security number that is used to identify a person
4. region_category Represents the region that a customer belongs to
5. membership_category Represents the category of the membership that a customer is using
6. joining_date Represents the date when a customer became a member
7. joined_through_referral Represents whether a customer joined using any referral code or ID
8. referral_id Represents a referral ID
9. preferred_offer_types Represents the type of offer that a customer prefers
10. medium_of_operation Represents the medium of operation that a customer uses for transactions
11. internet_option Represents the type of internet service a customer uses
12. last_visit_time Represents the last time a customer visited the website
13. days_since_last_login Represents the no. of days since a customer last logged into the website
14. avg_time_spent Represents the average time spent by a customer on the website
15. avg_transaction_value Represents the average transaction value of a customer
16. avg_frequency_login_days Represents the no. of times a customer has logged in to the website
17. points_in_wallet Represents the points awarded to a customer on each transaction
18. used_special_discount Represents whether a customer uses special discounts offered dd
19. offer_application_preference Represents whether a customer prefers offers
20. past_complaint Represents whether a customer has raised any complaints
21. complaint_status Represents whether the complaints raised by a customer was resolved
22. feedback Represents the feedback provided by a customer
23. churn_risk_score 0 or 1 Customer will Stay or Exit

Data Cleaning

in this dataset, we found some incorrect observations, for example in some numerical variables we found some negative values that shouldn't be negative. all incorrect values were replaced by NAs.

Explorative Data Analysis (EDA)

to Do an EDA, we made some visualizations using seaborn package in python and data studio. the result we got from this analysis is that membreship_category and Negative_feedback variables both have an important effect on the churn rate. to minimize churn, the company should for example improve the customer experience by minimizing ads on the website and improving customer service. check the report made using data studio here

Data Preparation for machine learning

In order to prepare the data we dealt first with missing values. For numerical variables, missing values were replaced by the median. On the other hand, missing values were replaced by the mode in the case of categorical variables.

secondly, data were split into two parts. 80% as a training set and 20% as a test set

finally, variables were scaled using the Z-score Normalization

Machine Learning

Models applied to this problem are Deep Neural Networks, Decision Tree, Random Forrest, And XGBoost. By comparing the models, we found that XGBoost performs better with a test accuracy equal to 93.75%