A novel construction method of FCM classifier models based on the KDTICM theories was proposed. The limitation and problems in the FCM classifier model are analyzed in this paper. KDTICM theories are introduced into the classifier model to solve these limitation and problems. The modified FCM classifier model consists of model structure, activation functions, inference rules, cognitive entropy and learning algorithms. The new model is suitable for protein secondary structure prediction. Compared with the classical classification algorithms, the new model not only shows a better classification performance, but also has powerful noise-immune ability which renders it robust.