This paper presents a study on churn prediction in the payment services industry. Churn prediction is crucial in preventing customers from switching to competition and also in keeping and developing relationships with the customers. Previous studies mainly focus on the problem of churn prediction for customers. Using machine learning algorithms, this study, for the first time, predicts churn rates for IoT devices and, thereby, generalizes the concept to the usage of devices. A dataset on POS devices provided by Token Financial Technologies-a company that aims to develop a churn prediction system-was used for the analyses. The methods Naive Bayes, Generalized Linear Model, Logistic Regression, Deep Learning, Decision Tree, Random Forest, and Gradient Boosted Trees were applied. The experimental results show that the best predictions are obtained by the algorithms Random Forest and Fast Large Margin. According to experimental results, battery life has a significant effect on the device churn as well as lifetime value of device has a major impact. The predictions helped Token Financial Technologies to save more than 60% of the usage of EFT-POS devices from potential churns by changing batteries and EFT-POS devices in the last quarter of 2019.