Abstract: Customer churn is one of the subscription-based business critical tasks that a company needs to make a decision about revenue stream management and to take care of their customers from churning. This research work is an introduction to a machine-learning method for customer churn analysis using predictive models. The process starts with a vast customer transaction dataset, which needs to be transformed into churn labels. Next, the system utilizes several machine learning algorithms, including logistic regression, decision tree, random forest, support vector machines (SVM), and gradient boosting, to process the input data and design predictive models. Carrying out feature selection and feature construction is part of the process. Feature selection is a method used to reduce the input of the dataset that might conflict with the output. Feature construction will unfortunately be a million-dollar question. Accuracy, precision, recall, and F1 scores measure model performances. In addition, the ROC curve can be obtained for a designed model.The findings demonstrate how accurate and efficient the proposed method can be for a customer churn problem. An organization gets an early warning about a customer churn problem using this method. It will put a customer in the retention consideration set. In addition, a design model gives an organization a reason behind a customer churn. This analysis will help organizations understand the cause of Churn and decide what they will do before a customer leaves.

Keywords: Customer Churn Prediction, Machine Learning, Predictive Modeling.

| DOI: 10.17148/IARJSET.2024.11708