Research Of Classification Problem Based On Fuzzy Logistic Neural Network

The fuzzy logistic neural network which combines the advantages of both fuzzy neural network and logistic regression, not only has a strong self-learning ability and ability to deal with prior knowledge of experts, but also enhances the interpretability for classification problems of neural network to a large extent. This paper introduces the binary classification model and its extended model of fuzzy logistic neural network separately, and the corresponding parameter identification algorithms. The main work of this paper is shown as follows.Firstly, the paper presents a hybrid learning algorithm to identify the parameters of fuzzy logistic neural network binary classification model. We choose Gaussian membership function as the membership function of the network antecedent, and use the constrained gradient descent algorithm to determine the center and width of Gaussian membership function. We choose logistic function as the activation function of the network subsequent, and use the maximum likelihood estimation method to identify the parameters of logistic function. Further, Gaussian particle swarm optimization (GPSO) algorithm is introduced to optimize the learning rate and the momentum factor of these two methods. The simulation results indicate that the binary classification model of fuzzy logistic neural network presented by this paper has a good performance on the UCI data sets and medical experimental data.Secondly, the binary classification model of fuzzy logistic neural network is extended to the kernel fuzzy logistic neural network by this paper to deal with multi-classification problem. The input data sets are mapped into a high dimensional feature space by linear kernel function and Gaussian kernel function, and data dimension reduction is accomplished by principal component analysis (PCA). On the basis of data preprocessing, maximum likelihood estimation method is used to identify the parameters. At the same time, we use the classification accuracy of test sample set to construct a stretch factor, which can improve the convergence speed and classification rate of the network by adjusting the learning rate dynamically. The simulation results indicate that the kernel fuzzy logistic neural network presented by this paper has some advantages on dealing with multi-classification UCI data sets. Furthermore, the use of variable universe learning rate also significantly improves the computational efficiency of the model.