Predicting Customer Churn using Neural Networks

Abstract

In today’s competitive telecom industry, retaining customers is as crucial as acquiring new ones. Customer churn—when a user discontinues service—is a major concern that can significantly impact a company's revenue. Traditionally, surveys were employed to identify dissatisfied customers, but these methods are labor-intensive and time-consuming. Some companies have adopted machine learning techniques for churn prediction; however, many of these models lack the accuracy needed for proactive retention strategies.
In this project, we propose a Neural Network-based approach to accurately predict customer churn. The model is enhanced with hyperparameter tuning to identify the optimal number of layers and neurons for improved performance. The dataset used is the publicly available "Customer Churn" dataset from Kaggle. Extensive data preprocessing steps—including normalization, categorical encoding, and exploratory data analysis—were conducted to prepare the data. The model's effectiveness is evaluated using metrics such as accuracy, precision, recall, F1-score, confusion matrix, and ROC curve. The final model achieved high prediction accuracy, demonstrating that deep learning combined with hyperparameter optimization can provide telecom companies with actionable insights to reduce customer attrition.