Telco customer churn is a dataset published on Kaggle that provides data about 7044 telecom subscribers. See how Visao makes it easy to explore the complex interactions between the 21 variables and identify which values contribute to churn.
dataset source: https://www.kaggle.com/blastchar/telco-customer-churn
Here is what the dataset looks like:
The dataset 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
The dataset consists of 7,044 customers and 21 variables:
- customerID: Customer ID
- gender: Whether the customer is a male or a female
- SeniorCitizen: Whether the customer is a senior citizen or not (1, 0)
- Partner: Whether the customer has a partner or not (Yes, No)
- Dependents: Whether the customer has dependents or not (Yes, No)
- tenure: Number of months the customer has stayed with the company
- PhoneService: Whether the customer has a phone service or not (Yes, No)
- MultipleLines: Whether the customer has multiple lines or not (Yes, No, No phone service)
- InternetService: Customer’s internet service provider (DSL, Fiber optic, No)
- OnlineSecurity: Whether the customer has online security or not (Yes, No, No internet service)
- OnlineBackup: Whether the customer has online backup or not (Yes, No, No internet service)
- DeviceProtection: Whether the customer has device protection or not (Yes, No, No internet service)
- TechSupport: Whether the customer has tech support or not (Yes, No, No internet service)
- StreamingTV: Whether the customer has streaming TV or not (Yes, No, No internet service)
- StreamingMovies: Whether the customer has streaming movies or not (Yes, No, No internet service)
- Contract: The contract term of the customer (Month-to-month, One year, Two year)
- PaperlessBilling: Whether the customer has paperless billing or not (Yes, No)
- PaymentMethod: The customer’s payment method (Electronic check, Mailed check, Bank transfer (automatic), Credit card (automatic))
- MonthlyCharges: The amount charged to the customer monthly
- TotalCharges: The total amount charged to the customer
- Churn: Whether the customer churned or not (Yes or No)
By uploading the dataset on Visao platform, the dataset is processed and a graph is created displaying the relations between the values of the 21 variables. The graph is built from the correlation between the variables (a link is displayed between variable values if its lift is > 1,1). The thickness of the link is proportional to the correlation.
The graph above shows that subscribers who are churning tend to, by decreasing influence, have a short term tenure (0-14 months), use the electronic check as payment method, be senior, have no partner, have low total charges (18,8-679) and high monthly charges (83,95-118,75), use paperless billing, and have no dependent. On the graph the width of the link between the 2 values is proportional to the strength of their relationship (measured by the difference to the theoretical independence, that we call the "lift" to simplify). By rolling over the link, the value of the strength is displayed.
Conversely, the graph above shows that subscribers who tend to, by decreasing influence, have a long tenure (48-72 months), use the mailed check/automatic bank transfer/credit card as payment method, have high total charges (2758-8685) and low monthly charges (18,25-50,35), use paperless billing, and have dependents.
We can investigate deeper the main explicable variables: tenure, payment methods, monthlyCharge. Low tenure (0-14 months) tends to be associated with variables contributing to churn (cluster of values on the right) or not churn (cluster on the left). It shows that people paying by mail check, have no multiple lines and low monthly charge, tend not to churn.
Looking at the variable MonthlyCharge, by increasing granularity on its discretisation (5 quantiles instead of 3), we can see that the top interval (94,25 - 118,75) relates to subscriptions that includes fiber optic, TV streaming, movies streaming, multiples lines, device protection, and online backup. The second highest monthyCharge (79,15 - 94,25), relates to fiber optic only, not the other options.
By further increasing granularity on the variable MonthlyCharge (10 deciles), we can see that 2 other options are added to the top interval (102,65 - 118,75): tech support and online security, whereas the top second interval (94,25 - 102,6) have fiber optic, TV streaming, and streaming movie, and the top third interval (85,55 - 94,25) fiber optic only. We could infer from these data in broad terms the tariff structure offered by this telco operator for each option.
Let's look at the type of options selected by clients depending on the total charge. The bottom deciles (bottom 1, 2, 3, 4, intervals 18,8 - 1403) are all clustered in the same group, where all options values are "no".
The bottom 5 (1401 - 2065) is the lowest decile to have a link with an option (with a lift of 1,24), i.e. multiple lines. This option must bring substantial value to these clients.
To get a global picture while keeping all variables in the graph, it is possible to select the links between variables with the highest correlations only. By requiring a minimum lift of 2,1 (instead of 1,1 by default), we can see that low monthly charges relates to clients choosing minimum service (no streaming TV, no online backup, nostreaming movies,...) and paying by mailed check. The variable churn disappeared because its lifts with other variables are too low.
Because Visao maps correlations between variables, it can identify divergences between datasets on the same variables, including for a predictive model the divergence between the dataset used for training and the dataset applied to new items.
We have used the dataset above on telecom customer churn to generate a model to predict whether a new customer will churn or not (based on a decision tree model). On the graph above, you can see the correlations between "churn - yes" and the various modalities, for the training dataset (on the left) compared to the predictive model (on the right). The lilft on the top quantile of monthly charge (versus churn=yes) has decreased from 1,27 to 0,98 and thefore dispappeared from the predictive model (on the right).
Visao measures the proportion of errors for each modality and provides the list of items likely to be false.
We could thus continue to provide more perspectives in this analysis but it would be more useful to do it with people with issues regarding churn and marketing. Do not hesitate to contact us to deepen this analysis, we will explain the various possible options for the valorization of the dataset, and the theoretical foundations of the various indices that Visao calculates.
