The following is an analysis of a telecom customer dataset found on Kaggle.com.
The data contains 5986 individual customer records with 19 substantial features. The response variable is Churn
, whether or not a customer has terminated their service with the telecom company.
Of the predictor variables, three are numeric: tenure
, MonthlyCharges
, and TotalCharges
. These variables are all related to the length of time and price of the service. The remaining 16 predictors are indicator variables covering customer demographics and the level of service provided by the company. A detailed description of the each feature can be found in the Feature table at the end of this document. A feature containing customer ID was removed as it was not relevant to this analysis.
Before attempting to predict customer churn we will examine the data with the intent to gain further insight into the following questions:
First let us look at the total amount of retained customers and churned customers:
Out of the 5986 observations, 4399 customers remained with the company while 1587 were churned. Overall, slightly over one-in-four, 26.5 percent, of customers terminated service at the end of the billing period.
Next we want to know the difference in monthly payments between the two groups:
Churn | Mean | FirstQuartile | Median | ThirdQuartile | StdDev |
---|---|---|---|---|---|
No | 61.42 | 25.12 | 64.75 | 88.7 | 31.09 |
Yes | 74.16 | 55.67 | 79.50 | 94.4 | 24.97 |
Customers who churn are paying on average $13 more dollars per month than those who do not. Seventy-five percent of customers who terminated service paid between $55.67 and $94.40. Further analysis is needed to see why this is the case.
Churn | Mean | FirstQuartile | Median | ThirdQuartile | StdDev |
---|---|---|---|---|---|
No | 2568.29 | 579.00 | 1689.45 | 4287.20 | 2335.46 |
Yes | 1550.70 | 131.93 | 706.60 | 2366.78 | 1905.71 |
Looking at total charges we can see that customers who churn seem to be paying higher monthly fees, but have a smaller amount of total charges before they churn. Tenure explains this by showing that the majority of churned customers are cancelling service within the first month before they accumlate large amounts of total fees.
Churn | Mean | FirstQuartile | Median | ThirdQuartile | StdDev |
---|---|---|---|---|---|
No | 37.60 | 15 | 37 | 61 | 24.07 |
Yes | 18.25 | 2 | 10 | 30 | 19.67 |
Since the data is skewed, the median is a better measure of central tendency than the mean. Half of the customers who churned have been with the company for 10 months or less, while the median retained customer has been with the company for 37 months.
Key Takeaways
A selection of demographic subgroups is listed below. The variable gender was omitted as it didn’t have a large effect on the the churn rate or average total charge.
Demographic Group | Churn Rate | Number of Customers | Average Total Charges |
---|---|---|---|
Overall | 0.265 | 5986 | $2298.06 |
All Non Senior citizens | 0.236 | 5020 | $2205.04 |
All Senior citizens | 0.416 | 966 | $2780.5 |
Senior citizens without Partners | 0.486 | 481 | $2038.3 |
Non Senior citizens with Partners | 0.169 | 2419 | $2962.68 |
Non Senior citizens with Dependents | 0.153 | 1718 | $2471.46 |
Non Senior citizens with Partners and Dependents | 0.14 | 1419 | $2695.63 |
Non Senior citizens without Partners and Dependents | 0.31 | 2302 | $1514.56 |
As shown in the table above, retirees had the highet churn rate. Almost 1 in 2 retirees without a partner churned; however the senior citizens had a higher than average total charges. Non-seniors with families spent more money on average and had lower churn rates. Non-seniors without families spent less money and had a 31% churn rate.
From this information, we can infer that those who have families are less likely to churn. While those who live alone are more likely to churn and will spend less money on average.
Key Takeaways
The features relating to contract infromation can be subdivided into these additional categories:
The type of service can be broken down into: Phone only, Internet Only, and Phone and Internet. The internet categories can be further divided into DSL and Fiber Optic. Fiber is generally more expensive than DSL.
Type of Service | Count | Churn Rate | Average Monthly Charge($) | Average Tenure |
---|---|---|---|---|
Phone Only | 1291 | 0.077 | 21.09 | 30.43 |
Internet Only DSL | 590 | 0.253 | 41.94 | 31.7 |
Internet Only Fiber | No Customers | 0 | 0 | 0 |
Phone & Internet DSL | 1478 | 0.168 | 64.56 | 33.21 |
Phone & Internet Fiber | 2627 | 0.415 | 91.55 | 33.23 |
The data shows that 41.5% of customers with Phone and Fiber Optic internet churned. This is most likely related to the average price of the fiber optic service at $91/month.
In the above graph the top line represents the lifetime of an average Fiber Optic internet customer, while the bottom line represents the lifetime of an average phone-only customer. Clearly customers using fiber optic for a short time or customers who pay more than the average fiber optic price are those who most frequently churn.
Key Takeaways
To analyze the number of extra service features I created a new variable numExtra
that collates the number of extra service features per each account. This variable decreases the amount of information so it will not be included a predictive model.
The highest churn rate is found among those with 0 to 2 extra features, while adding more than two features is associated with a decrease in the churn rate.
The churn rate breakdown for each feature is as follows:
Service | Number of Customers | Churn Rate |
---|---|---|
Multiple Lines | 2548 | 0.28 |
Online Security | 1713 | 0.14 |
Online Backup | 2090 | 0.22 |
Device Protection | 2055 | 0.27 |
Tech Support | 1735 | 0.16 |
Streaming TV | 2306 | 0.4 |
Streaming Movies | 2339 | 0.41 |
Key Takeaways
Contract Type | Number of Customers | Churn Rate |
---|---|---|
Month-to-month | 3269 | 0.43 |
One year | 1276 | 0.12 |
Two year | 1441 | 0.03 |
The data shows that most customers are on month-to-month contracts, and unsurprisingly customers who are on month-to-month contracts are more likely to churn.
Payment Method | Number of Customers | Churn Rate |
---|---|---|
Bank transfer (automatic) | 1308 | 0.17 |
Credit card (automatic) | 1303 | 0.15 |
Electronic check | 2006 | 0.45 |
Mailed check | 1369 | 0.19 |
A plurality of customers using electronic check as the payment method. This group is also the most likely to churn.
Paperless Billing | Count | Churn Rate |
---|---|---|
No | 2458 | 0.17 |
Yes | 3528 | 0.33 |
Key Takeaways
Predictor | Type | Description |
---|---|---|
Gender | Categorical | The gender of the customer (Male/Female) |
SeniorCitizen | Categorical | The retirement status of the customer (Yes/No) |
Partner | Categorical | Whether the customer is married(Yes/No) |
Dependents | Categorical | Whether or not the customer lives with dependents(Yes/No) |
Tenure | Discrete | The number of months a customer has been with the company |
PhoneService | Categorical | Connection status of telephone service (Yes/No) |
MultipleLines | Categorical | Status of clients multiple lines (Yes/No/No Telephone) |
InternetService | Categorical | Type of internet service (DSL/FiberOptic/None) |
OnlineSercuity | Categorical | Status of Online Security service (Yes/No/No Internet) |
OnlineBackup | Categorical | Status of Online Backup service (Yes/No/No Internet) |
DeviceProtection | Categorical | Status of device insurance service (Yes/No/No Internet) |
TechSupport | Categorical | Status of technical support service (Yes/No/No Internet) |
StreamingTV | Categorical | Status of streaming TV service (Yes/No/No Internet) |
StreamingMovies | Categorical | Status of movie streaming service (Yes/No/No Internet) |
Contract Type | Categorical | Type of contract (Month-to-month/One-Year/Two-Year) |
PaperlessBilling | Categorical | Paperless billing service (Yes/No) |
PayymentMethod | Categorical | Method of Payment (Bank Transfer(auto)/ Credit Card(auto)/Electronic Check/Mailed Check) |
MonthlyCharges | Continuous | The monthly payment made by the customer |
TotalCharges | Continuous | The total amount of money paid by the customer |