Machine Learning (ML) has become a huge topic in the field of Software Engineering in recent times. An important step in the development of a software system is the selection of a fitting reference architecture to base the system on. This also applies for software systems in the context of Machine Learning. The goal of this study was to collect reference architectures for ML-based systems and compare them based on a selected use case from the field of regression-based models. We performed a literature review to find reference architectures in scientific literature. We found some literature on the topic, from which we extracted six reference architectures. Some of the architectures are already well known and also used outside of ML topics. From the six architectures we found, we selected two based on their applicability for our use case and implemented a prototype system for each of them, both with the same functionality. Using all the information we gathered and our prototype systems, we then evaluated the benefits and drawbacks of the different architectures. We envision that our results can help practitioners in the process of choosing a reference architecture for their system.