In cognitive radio networks, the cooperation of different network entities holds great promise for attaining high-precision spectrum sensing decisions. It makes it possible for network users to cooperatively share sensing measurements to track the primary spectrum occupancy. However, this has made it easier for hostile secondary users (SUs) to launch a variety of attacks or interference into the system by sending false observations from local sensing. Our proposal is based on two techniques of detecting anomalies in spectrum data to inform whether there exists a malicious user (MU) or not. These techniques are the partial periodic pattern mining (PPPM) and the moving average time series (MATS) anomaly detection which compare the power allocation factors (PAFs) of the user signals both over time and in real time. An anomaly detection scheme combining the two techniques is presented and analyzed. The results indicate that combining the MATS and the PPPM techniques results in a balance between sensitivity to variations in PAF, specificity in detecting the true anomalies and reliability in anomaly detection. The proposed technique has good precision and F1 score performance when K = 2.