To thrive in today’s data-driven market, companies must master the art of consumer behaviour analysis. This study explores Federated Machine Learning (FML), a technique that enables machine learning on decentralized customer data, ensuring privacy. We compare the performance of various algorithms within FML for customer behavior analysis using a bank dataset. Our evaluation metrics include accuracy, precision, and log loss. Random Forest outperforms other algorithms, achieving the highest accuracy, precision, and lowest log loss, making it a reliable choice for customer behavior prediction in FML environments. This study paves the way for future research on scalability, real-world application testing, and the impact of federated learning protocols on model performance.