This project involves developing a machine learning-based system to predict customer churn in the telecommunications sector. The objective is to analyze historical customer data and build a classification model that identifies customers who may be at risk of discontinuing their services.
The project follows a structured workflow beginning with business problem identification, customer lifecycle understanding, data exploration, and data preprocessing. Exploratory Data Analysis includes univariate and bivariate analysis to examine feature distributions, identify relationships between variables, and understand patterns associated with customer churn.
The model development phase includes building a baseline classifier, experimenting with data balancing techniques such as SMOTE and SMOTEENN, and optimizing hyperparameters using Optuna. Candidate model configurations are evaluated using appropriate classification metrics to identify a suitable model for the prediction task.
The selected model is saved and integrated into a Streamlit-based web application. The application enables users to provide customer details and obtain model-generated churn predictions. The project concludes with application testing and deployment on AWS, demonstrating the practical implementation of an end-to-end machine learning solution.
Overall, the project demonstrates how data analysis, machine learning, class imbalance handling, hyperparameter optimization, and cloud deployment can be combined to develop a predictive analytics application that supports customer retention decisions.