Opportunistic usage of spectrum owned by licensed (or primary) users is the cornerstone on which the Cognitive Radio technology is built. Unlicensed (or secondary) users that thus use the spectrum rely opportunistically on spectrum sensing to determine the presence of primary user signal. In such a context, an attacker may mimic the behavior of a primary user (PU) to deceive the secondary users (SUs) into believing that a PU signal is present whereas it is not. Such an attack is known as the Primary User Emulation Attack (PUEA). A malicious user may launch a PUEA with the intention of grabbing the vacant bands for its own transmission. Another reason may be to simply disrupt the functioning of the Cognitive Radio Network (CRN). This work investigates the use of one-class classification for detecting PUEA in an infrastructure-based CRN. We opine that sensing data collected at the fusion center mainly for Collaborative Spectrum Sensing (CSS) can be exploited to characterize a PU signal. The PU signal features thus learned can aid in distinguishing a PU signal from a PU signal emulation. In particular, we investigate the use of one-class classification techniques, viz., Isolation Forest, Support Vector Machines (SVM), Minimum Covariance Determinant(MCD) and Local Outlier Factor(LOF) for detection of PUEA attacks. Simulation results support the validity of using one-class classification for detection of PUEA.