In cognitive radio network (CRN), secondary users (SUs) suffer from the spectrum sensing data falsification (SSDF) attack launched by malicious users (MUs). To deal with SSDF attack, one of the typical artificial neural networks (ANN): self-organizing map (SOM) neural network is recommended. SOM network possesses the ability of classifying the SUs into categories with different frequency of occurrence. Exploiting this characteristic, an algorithm calculating the suspicion degree (SD) of each SU is proposed. Considering that SD can't maintain the stability of classification results, we further propose the concept of average suspicion degree (ASD) and readjust the weights of SUs according to their ASD. Simulation results reveal that compared with the results of traditional SOM algorithm and dynamic threshold algorithm, the detection performance of our method shows improvement.