telco-report-project

Project: Find drivers for customer churn at Telco.

Project Goal

• Part I Find the drivers of customers churn.
• Part II Build the machine learning model that can predict customer's churn with accuracy > 75%
• Drow conclusions and define the next steps

Steps to Reproduce

1. Clone this repo into your computer.
2. Acquire the data from Codeup using your env.py file
3. Put the data in the file containing the cloned repo.
4. Run the telco_project.ipynb file.

Initial Thoughts

My initial hypothesis is that main churning factors are are connected with the monthly charges customers have and the time they have spent with the company.

The Plan

Part I

1. Acquire the data from the telco database. Transform the data to a Pandas data frame to make it easy to use and manipulate in the Jupyter Notebook.
2. Prepare the data for exploration and analysis. Find out if there are some values missing and find a way to handle those missing values.
3. Explore the data through visualizations and statistical tests. Find which features are connected with the customers' churn and which ones are not.
4. Make the exploration summary and leave the recommendation, where the company has to pay more attention

Part II

1. Pick the features that can help to build a good predicting model.
2. Pick the algorithms for creating the predicting model.
3. Encode the categorical variables
4. Split the data into 3 data sets: train, validate and test data (56%, 24%, and 20% respectively)
5. Create the models and evaluate them using the accuracy score on the train data sets.
6. Pick the models with the best accuracy score and evaluate them on the validation set.
7. Find out which model has the best performance: relatively high predicting power on the validation test and slight difference in the train and validation prediction results.
8. Apply the predictions to the test data set. Show the final accuracy scores. Save the predictions to the *.csv file

Drow conclusions and define the next steps

Data Dictionary

Feature	Definition	Values
Categorical Data
gender	The gender of the customer	'Female', 'Male'
senior_citizen	Gives the information if the customer is a senior citizen	0 -not senior, 1 - is senior citizen
partner	Shows if the customer has a partner	'Yes', 'No'
dependents	Information if the customer has dependents	'Yes', 'No'
Phone services
phone_service	Phone service connected	'Yes', 'No'
multiple_lines	Multiple phone lines connected	'Yes', 'No', 'No phone service'
Internet services
internet_service_type	Type of internet service customer has	'DSL', 'Fiber optic', 'None'
online_security	Online security included	'Yes', 'No', 'No internet service'
online_backup	Online backup included	'Yes', 'No', 'No internet service'
device_protection	Device protection included	'Yes', 'No', 'No internet service'
tech_support	Tech support included	'Yes', 'No', 'No internet service'
streaming_tv	Streaming TV included	'Yes', 'No', 'No internet service'
streaming_movies	Streaming movies	'Yes', 'No', 'No internet service'
Financial categories
paperless_billing	Does customer use paperless billing	'Yes', 'No'
contract_type	What type of contract has the customer	'One year', 'Month-to-month', 'Two year'
payment_type	How does the customer pay for the service	'Mailed check', 'Electronic check', 'Credit card (automatic)', 'Bank transfer (automatic)'
Numerical Data
monthly_charges	Monthly payment of the customer	Amount in USD
total_charges	Total amount paid to the company during all the period	Amount in USD
tenure	How long did the customer stay with the company	Number of months
Other Data
customer_id	Unique customer's id number
Target Data
churn	Did the customer churn?	'Yes', 'No'

Part I - Exploration

Takeaways and recommendations

1. Having or not having a phone service doesn't affect the customer’s churn much.
2. Fiber optic customers churn more.
   • Recommendation: research the competitor's market and reconsider the payment for the fiber optic internet or include more additional services for free for those customers.
3. Most of the churn happens within the first few months. My guess is that those customers are travelers that use the phone/internet services only for a short period of time.
   • Recommendation:
     • Create special attractive 'travelers packages’ and track them as well. This will help to identify if the churn is linked to travel reasons or if the company has something that makes new customers unhappy.
     • Offer some additional service discounts at the end of the first half-year period.
4. Most customers that churn have month-to-month contracts.
   • Recommendation: give more incentives for signing one-year or two-year contracts
5. Churned customers also had higher monthly payments. Monthly payments correlate with additional services. We can not recommend cutting the services, but we can recommend making more attractive service packages.
   • Recommendation: create more attractive service packages with better quality/price relationship

Part II - Modeling

• Almost all features in the data set significantly affect the customer's churn. My choice was to create feature combinations and test how various algorithms perform.
• Logistic Regression performs much better in terms of showing similar results with train, validate and test data sets. It won the first two positions in my 'best model' criterion.
• The 3rd best model was Desicion Tree with max_depth hyperparameter set to 3.
• Overall, my goal was to create a model that has a prediction power > 75%, and my 'winner' successfully coped with the task.

Conclusions and the next steps

• Most of the data presented in the telco data set affects the customer's churn. After exploring data I can name the biggest drivers of churn at the company, so the company can pay more attention to the customer that are high risk of churn. The drivers are:
• contract type,
• the internet service type,
• the payment type,
• monthly charges.

The model that have got the highest accuracy score included all of them.

Next steps The analysis provided in this project is just a first step on the way to deeper explorations and predictions. The next step in the exploration would be a multivariate analysis. Do the customers with dependents prefer Streaming Tv or Device protection? Do their preferences make them stay with the company or leave? Who are the people that sign Two-year contracts? Senior citizens, married couples? What services do they prefer to include? What contracts do fiber optic customers sign? What payment types do they use?

• The multivariate analysis will help to see the churn drivers better and possibly will help to build a better prediction model.