Usage profiles of a smart appliance predict how the machine is expected to interact with its users according to its usage history, but, the problem of building usage profiles has been scarcely discussed in the literature. In this paper, we discuss general aspects of generating usage profiles and propose a daily pattern based probability model for usage profiling. We show how the probability models can be learned with Bayesian network classifiers and we highlight the importance of finding the optimal days-of-the-week representation. An algorithm using the conditional log-likelihood minimum description length (CMDL) and hierarchical clustering is designed to find the representation. The learned model is then used in a Bayesian network classifier setting to predict usage profiles. The methodology is tested on a real-life dataset of office printers in a campus environment.