Multiple Time-Scale Pattern Classification Using a Refinement Fuzzy Min-Max Neural Network
Zheng Li, Yibing Han, Xiaolong Wu · Preprints.org · 2024
This paper proposes a multiple time-scale pattern classification method based on a refinement fuzzy min-max neural network (RFMMNN). The purpose is to provide the suitable hyperboxes for RFMMNN to cover the multiple time-scale input patterns in the online learning algorithms. Firstly, a new fuzzy production rule (FPR) with local and global weights is established based on the multi-time scale input pattern. This FPR can directly use multi-scale features for pattern classification. Secondly, a fuzzy min-max network (FMM) with an enhanced learning algorithm is developed, and the FMM is used to refine the local and global parameters of FPR. Thirdly, a pruning strategy is designed to prevent the useless FPR generation in the learning process, and further to improve the accuracy of classification. The efficacy of FFMMNN is evaluated using benchmark data sets and a real-world task. The results are better than those from various FMM-based models, such as support vector machine-based, Bayesian-based, decision tree-based, fuzzy-based, and neural-based classifiers.