Skip to content
Related Articles

Related Articles

Python | Customer Churn Analysis Prediction

Improve Article
Save Article
  • Difficulty Level : Medium
  • Last Updated : 23 Mar, 2020
Improve Article
Save Article

Customer Churn
It is when an existing customer, user, subscriber, or any kind of return client stops doing business or ends the relationship with a company.

Types of Customer Churn –

  • Contractual Churn : When a customer is under a contract for a service and decides to cancel the service e.g. Cable TV, SaaS.
  • Voluntary Churn : When a user voluntarily cancels a service e.g. Cellular connection.
  • Non-Contractual Churn : When a customer is not under a contract for a service and decides to cancel the service e.g. Consumer Loyalty in retail stores.
  • Involuntary Churn : When a churn occurs without any request of the customer e.g. Credit card expiration.

Reasons for Voluntary Churn

  • Lack of usage
  • Poor service
  • Better price

Code: Importing Telco Churn dataset

# Import required libraries
import numpy as np
import pandas as pd
# Import the dataset
dataset = pd.read_csv('telcochurndata.csv')
# Glance at the first five records
# Print all the features of the data


Exploratory Data Analysis on Telco Churn Dataset

Code : To find the number of churners and non-churners in the dataset:

# Churners vs Non-Churners


Code: To group data by Churn and compute the mean to find out if churners make more customer service calls than non-churners:

# Group data by 'Churn' and compute the mean
print(dataset.groupby('Churn')['Customer service calls'].mean())


Yes! Perhaps unsurprisingly, churners seem to make more customer service calls than non-churners.

Code: To find out if one State has more churners compared to another.

# Count the number of churners and non-churners by State


While California is the most populous state in the U.S, there are not as many customers from California in our dataset. Arizona (AZ), for example, has 64 customers, 4 of whom ended up churning. In comparison, California has a higher number (and percentage) of customers who churned. This is useful information for a company.

Exploring Data Visualizations : To understand how variables are distributed.

# Import matplotlib and seaborn
import matplotlib.pyplot as plt
import seaborn as sns
# Visualize the distribution of 'Total day minutes'
plt.hist(dataset['Total day minutes'], bins = 100)
# Display the plot


Code: To visualize the difference in Customer service calls between churners and non-churners

# Create the box plot
sns.boxplot(x = 'Churn',
            y = 'Customer service calls',
            data = dataset,
            sym = "",                  
            hue = "International plan"
# Display the plot


It looks like customers who do churn end up leaving more customer service calls unless these customers also have an international plan, in which case they leave fewer customer service calls. This type of information is really useful in better understanding the drivers of churn. It’s now time to learn about how to preprocess your data prior to modelling.

Data Preprocessing for Telco Churn Dataset

Many Machine Learning models make certain assumptions about how the data is distributed. Some of the assumptions are as follows:

  • The features are normally distributed
  • The features are on the same scale
  • The datatypes of features are numeric

In telco churn data, Churn, Voice mail plan, and, International plan, in particular, are binary features that can easily be converted into 0’s and 1’s.

# Features and Labels
X = dataset.iloc[:, 0:19].values
y = dataset.iloc[:, 19].values # Churn
# Encoding categorical data in X
from sklearn.preprocessing import LabelEncoder
labelencoder_X_1 = LabelEncoder()
X[:, 3] = labelencoder_X_1.fit_transform(X[:, 3])
labelencoder_X_2 = LabelEncoder()
X[:, 4] = labelencoder_X_2.fit_transform(X[:, 4])
# Encoding categorical data in y
labelencoder_y = LabelEncoder()
y = labelencoder_y.fit_transform(y)

Code: Encoding State feature using One hot encoding

# Removing extra column to avoid dummy variable trap
X_State = pd.get_dummies(X[:, 0], drop_first = True)
# Converting X to a dataframe
X = pd.DataFrame(X)
# Dropping the 'State' column
X = X.drop([0], axis = 1)
# Merging two dataframes
frames = [X_State, X]
result = pd.concat(frames, axis = 1, ignore_index = True)
# Final dataset with all numeric features
X = result

Code : To Create Training and Test sets

# Splitting the dataset into the Training and Test sets
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, 
                                                    test_size = 0.2
                                                    random_state = 0)

Code: To scale features of the training and test sets

# Feature Scaling
from sklearn.preprocessing import StandardScaler
sc = StandardScaler()
X_train = sc.fit_transform(X_train)
X_test = sc.transform(X_test)

Code: To train a Random Forest classifier model on the training set.

# Import RandomForestClassifier
from sklearn.ensemble import RandomForestClassifier
# Instantiate the classifier
clf = RandomForestClassifier()
# Fit to the training data, y_train)

Code : Making Predictions

# Predict the labels for the test set
y_pred = clf.predict(X_test)

Code: Evaluating Model Performance

# Compute accuracy
from sklearn.metrics import accuracy_score
accuracy_score(y_test, y_pred)


Code : Confusion Matrix

from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_test, y_pred))


From the confusion matrix, we can compute the following metrics:

  • True Positives(TP) = 51
  • True Negatives(TN) = 575
  • False Positives(FP) = 4
  • False Negatives(FN) = 37
  • Precision = TP/(TP+FP) = 0.92
  • Recall = TP/(TP+FN) = 0.57
  • Accuracy = (TP+TN)/(TP+TN+FP+FN) = 0.9385

My Personal Notes arrow_drop_up
Related Articles

Start Your Coding Journey Now!