Interface unit cells often contain several molecules. Just from looking at the geometries, it is often hard to tell whether a given structure is just a defect of a certain thermodynamic phase (i.e., a misaligned molecule, a vacancy, etc.), or whether it is a new phase altogether. The target of this thesis is to test different parameters (e.g. bond-order ordering parameters) and to write a machine-learning based clustering algorithm to classify the different structures obtained by SAMPLE. This will allow us not only to predict the prevalence of defects in structures, but also whether a given phase transformation is a first-order or second order transition.