Accelerated fuzzy min–max neural network and arithmetic optimization algorithm for optimizing hyper-boxes and feature selection

• Published in: Neural computing & applications Vol. 36; no. 4; pp. 1553 - 1568
• Main Authors: Alzaqebah, Malek, Ahmed, Eman A. E.
• Format: Journal Article
• Language: English
• Published: London Springer London 01-02-2024; Springer Nature B.V
• Subjects: Algorithms; Arithmetic; Artificial Intelligence; Artificial neural networks; Boxes; Classification; Complexity; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Datasets; Fuzzy logic; Image Processing and Computer Vision; Neural networks; Optimization; Optimization algorithms; Original Article; Probability and Statistics in Computer Science; Search methods; Fuzzy minimum maximum; Feature selection; Arithmetic optimization algorithm; Hyperbox optimization
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-023-09131-6

The fuzzy min–max (FMM) neural network effectively solves classification problems. Despite its success, it has been observed recently that FMM has overlapping between hyper-boxes in some datasets which certainly the overall classification performance, as well as FMM has a high compactional complexity, especially when dealing with high-dimensional datasets. a hybrid model combining Arithmetic Optimization Algorithm (AOA) and Accelerated fuzzy min–max (AFMM) neural network is proposed to produce an AFMM-AOA model, where AFMM is used to speed up the hyper-boxes contraction process and to reduce the number of hyper-boxes, then AOA is employed for selecting the optimal feature set in each hyper-box, which results in lowering the compactional complexity and overcoming the overlapping problem. Furthermore, the AOA algorithm has been modified (MAOA) to enhance the exploiting ability of the original AOA algorithm for handling the high dimensionality in hyper-box representation by introducing both random and neighbor search methods. The performance of the proposed methods is evaluated using twelve datasets, as a result, the neighbor search method shows better performance than the random search. In addition, both methods showed superior performance compared with the original AOA and some state-of-the-art algorithms.