A NOVEL APPROACH TO SYMBOLIC DATA CLUSTERING USING ENHANCED K-MEANS ALGORITHM

Abstract

Clustering is a crucial technique in image analysis, yet traditional methods such as K-Means often struggle when dealing with complex, high-dimensional, or uncertain data. This limitation reduces their effectiveness in accurately grouping images, particularly when variability and overlapping features exist across categories. To address this problem, this paper introduces a novel approach that integrates symbolic data with the K-Means algorithm to cluster image data more effectively. By symbolically representing both color intensity and spatial features, we enhance the algorithm’s ability to handle variability and uncertainty. We test our method on the CIFAR-10 dataset, where it achieves a clustering accuracy of 94.0% with an Adjusted Rand Index of 0.7, outperforming traditional methods such as K-Means (82.5%), DBSCAN (78.1%), and Hierarchical clustering (81.3%). Our results demonstrate that symbolic data analysis offers a more flexible and accurate solution for image clustering, with potential applications in fields such as medical image processing and environmental monitoring. Limitations and directions for future research are also discussed.