The partition of urban networks for the application of aggregated traffic models based on the Macroscopic Fundamental Diagram (MFD) is a challenging task and an active research question in the literature. The partitioning of urban networks should yield fully connected and compact regions, ensuring the homogeneity of traffic conditions within each of them. This requires having rich datasets of traffic data available, which can be difficult to gather. Moreover, one should also decide the optimal number of regions to partition urban networks. Several studies have addressed some of these research questions; but, the models in the literature fail to ensure all the required properties of a good partitioning. In this paper, we propose to use Gaussian Mixture Models to partition urban networks. The covariance matrix allows the model to learn the topological features and dependencies of the urban network. We also discuss proxies that can be utilized to ensure the homogeneity of traffic conditions in the regions, when traffic data is not available.