TELECOM CUSTOMER CHURN PREDICTION USING MACHINE LEARNING

For telecommunication companies, losing customers can be particularly important since it can impact revenue
and future customer relationships. Determining customers who are at risk of leaving early can help companies
prevent their customers from leaving them. This project is a machine learning system for predicting customer
churn amongst telecom service providers based on the demographic, account, and service related information. It
is based on the Telco Customer Churn dataset provided by IBM which has 33 attributes and 7,043 customers
records. Unnecessary fields, like customer identifiers and geographic data, were stripped of from the data during
preprocessing, and variables that might lead to data leakage (like churn score, churn value, and churn reason) were
also removed. All the other data was cleaned with the process of missing value handling and one-hot coding of
categorical variables. Numerical features were standardized using Standard Scaler. Logistic Regression was
chosen as the main classification model and trained with 80:20 train-test split. The model was tested against 1,409
test customer records and it performed with an accuracy of 80.3%, precision of 67.9%, and a recall of 57.8% on
the churn class. The model also achieved an ROC-AUC of 0.854, which indicates that the model has a good
capacity to distinguish between customers who are likely to churn and those who are likely to stay. Furthermore,
5-fold stratified cross validation was used to assess the performance of the model, which yielded a mean accuracy
of 81.09% and a mean ROC-AUC of 0.858, supporting the performance of the model. The trained model, scaler,
and feature information were saved using Joblib to be used practically in the Streamlit web application. The
application enables the user to input the customer information, and instantly gets a probability of churn along with
a risk indicator. Various functional, interface, edge case, backend and security scenarios were tested. Churn
probabilities of 0.58 and 0.01 were returned for example customer profiles representing higher and lower risk
customers respectively. The project provides a valuable demonstration of the integration of machine learning, data
preprocessing techniques, and an interactive interface to create a practical decision support system for telecom
customer retention. The Customer Churn Prediction app is a machine learning application that predicts the churn
for customers in the telecom industry. Customer Churn Prediction app is a machine learning application for
predicting customer churn in telecom sector.

Customer Churn Prediction, Machine Learning, Logistic Regression, Feature Engineering, Data Preprocessing,