Beta-lactamase is a protein which is produced in bacteria and is a primary cause of antibiotic resistance to Beta Lactam antibiotics. Beta Lactams are among the most common types of antibiotics (including penicillin) and thus understanding the resistance conferred to bacteria by beta-lactamase enzymes is a critical step in developing effective drugs. There are many mutants of the beta-lactamase enzyme, each with the same function but different substrate affinities to the various types of antibiotics. We are interested in identifying intrinsic dynamics of apo structures, if any, to elucidate differences in substrate specificity, specifically the motions which can explain how Extended Spectrum beta-lactamases (ESBLs) are able to bind to such a wide variety of substrates. To characterize dynamical differences between mutant proteins, we are employing Essential Dynamics, which is the most commonly employed method to analyze the internal dynamics of proteins, as well as several other supervised machine learning methods. We show that Quadratic Discriminant Analysis, when used with a filtering method to circumvent the multicollinearity problem, can be used to classify structures from the wild type (TEM-1) and ESBL (TEM-52) with greater accuracy than obtained from unsupervised learning. Finally, an in-house supervised learning method is used to identify differences in molecular motion to elucidate mechanisms likely responsible for, at least in part, the experimentally determined differences in binding affinity between the ESBL TEM-52 with respect to the wild type TEM-1.