/Telco_Project

Repository for my Classification Project Modeling on the Telco_Churn Dataset.

Primary LanguageJupyter NotebookGNU Affero General Public License v3.0AGPL-3.0

Telco Project

Project Description

  • Analyzing Customer Churn Factors for Predictive Insights

The Telco Telecommunication company provides phone, internet, streaming and other add-on services to their customers. This project will involve analysis of the various elements of customer churns to determine if they increase or decrease the probability of customer churn

Project Goal

  • Identify drivers for churn of Telco customers
  • Utilze drivers to develop a Machine Learning model to classify churn as a customer ending their contract or not ending their contract with Telco.
  • The details of could provide further insight on which customer elements contribute to or detract from a customer churning.

Initial Thoughts

My initial hypothesis is that the drive of churn may be impacted by customers not opting for online security as part of their contract services.

The Plan

  • Acquire data from Codeup MySQL Database

  • Prepare data

    • Feature Engineer columns from existing data
      • online_security
      • online_backup
      • device_protection
      • tech_support
      • streaming_tv
      • streaming_movies
      • tech_support

  • Explore data in search of impactful drivers of churn

    • Answer the following initial questions,
      • Is Churn independent from online Security?
      • Is Churn independent from internet service types?
      • Is there a difference in churn based on monthly charges? Total charges?
      • Does having more than one add-on affect churn?
      • Does the contract type affect churn?

  • Develop a Model to predict if a customer will churn

    • Use drivers identified in explore to help build predictive models of different types
    • Evaluate models on train and validate data
    • Select the best model based on highest accuracy
    • Evaluate the best performing model on test data

  • Draw conclusions

Data Dictionary

Feature Values Definition
customer_id Alpha-numeric Unique ID for each customer
gender Female/Male Gender of customer
senior_citizen True=1/False=0 Whether customer is a senior citizen or not
partner True=1/False=0 True=1/False=0, whether customer has a partner or not
dependents True=1/False=0 True=1/False=0, whether customer has dependents or not
phone_service True=1/False=0 True=1/False=0, whether customer has phone service or not
multiple_lines Yes/No/No phone service Whether customer has multiple lines or not
internet_service_type None/DSL/Fiber Optic Which internet service customer has
online_security Yes/No/No internet service Whether customer has online_security
online_backup Yes/No/No internet service Whether customer has online_backup
device_protection Yes/No/No internet service Whether customer has device_protection
tech_support Yes/No/No internet service Whether customer has tech_support
streaming_tv Yes/No/No internet service Whether customer has streaming_tv
streaming_movies Yes/No/No internet service Whether customer has streaming_movies
contract_type 3 options Month-to-Month/One-year/Two-year, term of contract
payment_type 4 options (2 auto) Customer payment method
paperless_billing True=1/False=0 Whether a customer has paperless billing enabled
monthly_charges Numeric USD Amount customer is charged monthly
total_charges Numeric USD Total amount customer has been charged
tenure Numeric Number of months customer has stayed
churn (target) True=1/False=0 Whether or not the customer has churned
Additional Features True=1/False=0 Encoded values for categorical data

Steps to Reproduce

  1. Clone this repo following the code link at the top
  2. If you have granted access to the Codeup MySQL DB:
    • Save env.py in the repo with user, password, and host variables and add to .gitignore file
    • Run jupyter notebook or VS Code file
  3. If you don't have granted access:
    • Request access from Codeup staff
    • Return to step 2

Conclusions

Takeaways and Key Findings

  • Phone service was found to be a driver of churn.
  • Internet service was not a driver of churn.
  • Monthly charges was found to be a driver of churn, higher charges for churn.

Recommendations

  • Check with the finance department to figure out if there are issues with the phone service option.
  • Check to see if multiple phone lines have a impact on churn.

Next Steps

  • Given more time I could check what is causing the phone service leading to high churn.