Customer churn prediction is one of the most critical problems an e-commerce firm faces, particularly in the B2C segment, due to increasing customer acquisition cost. In this regard, the paper proposes a novelty churn prediction model using XGBoost as a supervised learning technique based on the gradient boosting algorithm to find out who all the potential customers that can be identified as churning and strategies can be provided to retain them. A model uses the myriad amount of customer data created at e-commerce businesses, such as searches, purchases, reviews, and comments, to analyse customer behaviour in searching for potential opportunities of attrition. For execution speed and performance of the model, XGBoost will be used, which allows big datasets to be processed without restriction. The proposed algorithm creates a series of models and then combines them into an overall model which is better than the individual model in the sequence. The performance of the proposed model looks very promising in comparison with other models of churn prediction based on accuracy and Kappa metrics. Concerning this, an accuracy rate of 97.80% and a Kappa score of 0.919 gave the XGBoost model the highest rank for enabling analysts to identify lost customers and develop retention strategies.