This is the code for a double-layer feedforward neural network that predicts which customers are at the highest risk of leaving a (simulated) bank company.
I built the model using the Keras library, which is built on top of Tensorflow and Theano. The inputs are a mixture of numeric, binary, as well as categorical values, and the output is binary (indicating whether the customer leaves the bank). I used adam for stochastic optimization, and binary crossentropy as the loss function.
- tensorflow
- keras
- numpy
- pandas
- scikit-learn
Install dependencies using pip.
I used a simulated dataset (input/churn.csv) with 10,000 observations (customers) and 13 attributes.
| Variable | Definition |
|---|---|
| CustomerId | Customer's account ID |
| Surname | Customer's surname |
| CreditScore | Customer's credit score |
| Geography | Country (France/Germany/Spain) |
| Gender | Customer's gender (Male/Female) |
| Age | Customer's age |
| Tenure | Number of years customer has been with the bank |
| Balance | Customer's account balance |
| NumOfProducts | Number of bank products used by customer |
| HasCrCard | Does customer have a credit card? |
| IsActiveMember | Is the customer an active member? |
| EstimatedSalary | Customer's estimated salary |
| Exited | Did the customer leave the bank? |
Run python script.py in terminal to see the network in training.
...
Epoch 98/100
8000/8000 [==============================] - 2s - loss: 0.3924 - acc: 0.8361
Epoch 99/100
8000/8000 [==============================] - 2s - loss: 0.3931 - acc: 0.8374
Epoch 100/100
8000/8000 [==============================] - 2s - loss: 0.3929 - acc: 0.8367
Training complete!
Testing complete!
Model accuracy: 0.8455
----------------
NEW OBSERVATION:
----------------
Geography: France
Credit Score: 600
Gender: Male
Age: 40 years old
Tenure: 3 years
Balance: $60000
Number of Products: 2
Does this customer have a credit card ? Yes
Is this customer an Active Member: Yes
Estimated Salary: $50000
-----------------------------------------------------------------------
Probability that new customer will leave the bank: 0.058069657534360886
-----------------------------------------------------------------------