Multicodebook Neural Network Using Intelligent K-Means Clustering Based on Histogram Information for Multimodal Data Classification

Abstract

Multimodal distribution makes classification more challenging. In some cases, it will cause a decrease in performance of the classifier. Therefore, we must select and utilize an appropriate classifier for this kind of data. In addition, multimodal distribution causes classification result of single-codebook-based classifiers fail to classify the data accurately. Some research has been conducted to build multicodebook-based classifier. In this study, we propose an enhancement of Multicodebook Fuzzy Neuro Generalized Learning Vector Quantization method using Intelligent K-Means clustering based on Histogram Information. First, We used Intelligent K-Means Clustering using histogram information as the number of cluster parameter. Next, We applied the clustering method to FNGLVQ in order to generate codebook candidate before the training process. The experiment is conducted using synthetic and benchmark datasets. Overall, the proposed method achieved a better accuracy compared to multi codebook FNGLVQ using IK-Means clustering based on anomalous pattern and original FNGLVQ with a margin of 4.14% and 17.29% respectively.