Publicação
pseudo-FNN
| Resumo: | This study presents the pseudo-FNN, a fuzzy neural network model that integrates the pseudo-unineuron, a novel neuron type leveraging pseudo-uninorms to enhance non-commutative operations and knowledge extraction. The pseudo-FNN employs a three-layer architecture with Gaussian fuzzy neurons, where weights are derived from kernel density estimation and rule consequents are optimized using multiple algorithms. Experimental evaluations on four datasets (Iris, Haberman, Transfusion, and Mammographic Masses) demonstrate the model’s competitive performance. The pseudo-FNN outperformed traditional fuzzy neural networks such as ANFIS and showed comparable results with optimization-enhanced FNNs. Among the optimization techniques, models using SGD, Adam, and RMSProp achieved the most consistent and high accuracies across datasets with pseudo-FNN models often aligning with these trends. Statistical analysis confirmed significant improvements over non-optimized models, and the pseudo-FNN demonstrated robustness in addressing varying classification complexities. These results highlight the effectiveness of the pseudo-unineuron in advancing fuzzy neural network architectures. |
|---|---|
| Autores principais: | de Campos Souza, Paulo Vitor |
| Assunto: | Fuzzy Neural Networks pseudo-uninorm pseudo-FNN |
| Ano: | 2026 |
| País: | Portugal |
| Tipo de documento: | artigo |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This study presents the pseudo-FNN, a fuzzy neural network model that integrates the pseudo-unineuron, a novel neuron type leveraging pseudo-uninorms to enhance non-commutative operations and knowledge extraction. The pseudo-FNN employs a three-layer architecture with Gaussian fuzzy neurons, where weights are derived from kernel density estimation and rule consequents are optimized using multiple algorithms. Experimental evaluations on four datasets (Iris, Haberman, Transfusion, and Mammographic Masses) demonstrate the model’s competitive performance. The pseudo-FNN outperformed traditional fuzzy neural networks such as ANFIS and showed comparable results with optimization-enhanced FNNs. Among the optimization techniques, models using SGD, Adam, and RMSProp achieved the most consistent and high accuracies across datasets with pseudo-FNN models often aligning with these trends. Statistical analysis confirmed significant improvements over non-optimized models, and the pseudo-FNN demonstrated robustness in addressing varying classification complexities. These results highlight the effectiveness of the pseudo-unineuron in advancing fuzzy neural network architectures. |
|---|