Detection and characterization of newly synthesized cannabinoids (NSCs) is challenging due to the lack of availability of reference standards and chemical data. In this study, a binary classification system was developed and validated using partial least square discriminant analysis (PLS-DA) by utilizing readily available mass spectral data of known drugs to assist in the identification of previously unknown NCSs. First, a binary classification model was developed to discriminate cannabinoids and cannabinoid-related compounds from other drug classes. Then, a classification model was developed to discriminate classical (THC-related) from synthetic cannabinoids. Additional models were developed based on the most abundant functional groups including core groups such as indole, indazole, azaindole, and naphthoylpyrrole, as well as head and tail groups including 4-fluorobenzyl (FUB) and 5-Fluoropentyl (5-F). The predictive ability of these models was tested via both cross-validation and external validation. The results show that all models developed are highly accurate. Additionally, latent variables (LVs) of each model provide useful mass to charge (m/z) for discrimination between classes, which further facilitates the identification of different functional groups of previously unknown drug molecules.