Yazar "Wang, Jian" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim Yok
    Öğe
    Adaptive Nonstationary Fuzzy Neural Network
    (Elsevier, 2024) Chang, Qin; Zhang, Zhen; Wei, Fanyue; Wang, Jian; Pedrycz, Witold; Pal, Nikhil R.
    Fuzzy neural network (FNN) plays an important role as an inference system in practical applications. To enhance its ability of handling uncertainty without invoking high computational cost, and to take variations in rules into consideration as well, we propose a new inference framework-nonstationary fuzzy neural network (NFNN). This NFNN is composed of a series of zero -order TSK FNNs with the same structure but using slightly perturbed fuzzy sets in the corresponding neurons, which is inspired from the non -stationary fuzzy sets and can mimic the variation in human's decision -making process. In order to obtain a concise and adaptive rule base for NFNN, a modified affinity propagation (MAP) clustering method is proposed. The MAP can determine the number of rules in an adaptive manner, and is used to initialize the rule parameters of NFNN, which we call Adaptive NFNN (ANFNN). Numerical experiments have been carried out over 17 classification datasets and three regression datasets. The experimental results demonstrate that ANFNN exhibits better accuracy, generalization ability, and fault -tolerance ability compared with the classical type -1 fuzzy neural network. In 15 of the 17 classification datasets, ANFNN achieves the same or better accuracy performance compared to interval type -2 FNNs with about half time consumed. This work confirms the feasibility of integrating simplestructured type -1 TSK FNNs to achieve the performance of interval type -2 FNNs, and proves that ANFNN can be a more accurate and reliable alternative to classical type -1 FNN.
  • Küçük Resim Yok
    Öğe
    A New Oversampling Method Based on Triangulation of Sample Space
    (Ieee-Inst Electrical Electronics Engineers Inc, 2024) Chen, Yueqi; Pedrycz, Witold; Wang, Jian; Zhang, Chao; Yang, Jie
    Coping with imbalanced data is a challenging task in practical classification problems. One of effective methods to solve imbalanced problems is to oversample the minority class. SMOTE is a classical oversampling method. However, it exhibits two disadvantages, namely, a linear generation and overgeneralization. In this article, an improved synthetic minority oversampling technique (SMOTE) method, FE-SMOTE, is proposed based on the idea of the method of finite elements. FE-SMOTE not only overcomes the above two disadvantages of SMOTE but also can generate samples that are more in line with the density distribution of the original minority class than those generated by the existing SMOTE variants. The originality of the proposed method stems from constructing a simplex for every minority sample and then triangulating it to expand the region of synthetic samples from lines to space. A new definition of the relative size for triangular elements not only helps determine the number of synthetic samples but also weakens the adverse impact of outliers. Generated samples by FE-SMOTE can effectively reflect the local potential distribution structure arising around every minority sample. Compared with 16 commonly studied oversampling methods, FE-SMOTE produces promising results quantified in terms of G-mean, AUC, F-measure, and accuracy on 22 benchmark imbalanced datasets and the big dataset MNIST.