So that's why Telecom customers jump tanks..

Monthly contract, only internet and Bill shock could be the doing the damage

Telecom industry is highly competitive. Retaining customer is challenge in itself. Understanding the factors behind customer behavior behind dumping one provider to avail competitors service, is key to retaining customers and to run the telecom services profitability.

So What are the primary cause for customer to leave..

Month-to-Month payment contract doesn't work for TeleCOs

Yes, you read it right. Data reveals that when the customer avails month on month payment contract mode , he or she is 42% more likely to move to another service provider than a customer with yearly or two years contract.

Only Fiber Optics Plan is not good enough

Next customers with Fiber for internet are mostly likely to leave, about 42% customers who use only fiber optics service are likely to churn out.

Bill Payment through Electronic Check 

Customers who have opted to pay bills using electronic check clearly have very high churn rate of upto 45%

Longer the customers stays the better 

The data shows a general tendency of the longer a customers stays with the Telecom service they is more likely to stay even longer. Similarly the longer stay means more money the customer has paid to the company is more likely to stay. It is interesting to note that if the customer has spent more that $8K with the telecom operator the chances of customer migrating is almost nil.

Monthly Charges are from $70 to $100 .. 

Look out for such customers .. well they can be on their way out.

There other factors which also shows high chances of leaving like, People who only use Internet without phone or other value added services like online backup or technical support.

It boils down to a deadly combination that could be summaried as

There is more than 70% of chance for a person using only Fiber optics internet service with month-on-month contract of $70 - $100 paying through Electronic Check leaving in a month.

The graph below shows the counts of subscribers who left versus who stayed against each feature. This give a broad picture of which feature contributes to higher churn rate.

Tenure and Monthly Charges

The data speaks for itself.

Interestingly, gender, having dependents or partners does not seem to have a significant impact on the customer churn.

What can be do to prevent customers leaving 

Luckily, there are machine learning model which can be build and trained to predict if the customer have a probability of leaving. This gives the Telecom companies an opportunity to proactively address any customer issue before it is too late. There are many ideas present in other article like here and data models describe in other articles used to predict Telcom customer churn.

With the data and models I have used to predict the probability of customer moving on, the Machine learning model using Logistic regression technique proves to be the best.

For those of are not aware of machine learning model or think it a machine with the intelligent of human being then check out here.

The chosen Model

There are many other ML model which could perform better but, the models I have used are 

1. Logistic regression using number encoding of feature

2. Logistic regression using dummy variable encoding of feature.

3. Random Forrest Ensemble. 

The metrics of the models used are compared here where Logistic regression with dummy variable substitution technique has better model metrics and can be used for prediction on new customers data

Logistic Regression model scores

LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True, intercept_scaling=1, max_iter=100, multi_class='ovr', n_jobs=1, penalty='l2', random_state=42, solver='liblinear', tol=0.0001, verbose=0, warm_start=False)

Classification report for dataset with Dummy variable replacement: precision recall f1-score support

      0       0.85      0.90      0.87      1539
      1       0.68      0.57      0.62       574

avg / total 0.80 0.81 0.81 2113

Accuracy Score : 0.811168954094 Area under curve : 0.737099071074

Hope this was useful. Thanks for your time.