Automatic bioswarm clustering

Abstract

Data clustering algorithms divide data into meaningful clusters so that the patterns in the same group are similar in some way, and the patterns in different clusters differ in the same way. The search for clusters involves unsupervised learning, in which it is not known in advance to which group the data belong. The partitioning group attempts to decompose the data into a set of disjoint groups through a criterion function that involves minimizing some measure of non-similarity of the objects within each cluster while maximizing the non-similarity of the different groups. The clustering problem can be seen as an optimization problem that locates the optimal centroids of clusters rather than finding the optimal partition. Most clustering algorithms require that the number of clusters be provided as a parameter for the solution. However, in real-world problems this value is not previously known. In automatic clustering, identifying the optimal number of clusters is part of the problem. The discovery of knowledge, including the task of clustering, is one of the great challenges of the Analytics in the Internet of Things. In this project, six Swarm Intelligence algorithms are used for automatic clustering and applied to numerical data sets. These algorithms will be developed to find the ideal number of clusters and their corresponding solution (coordinates of the centroids), through the optimization of division criteria based on measures of the quality of the clusters. Particle Swarm Optimization, Wolf Wolf Optimization, Enhanced Fish School Search, Algorithm Whale Optimization, Cuckoo Search, and Cat Swarm Optimization will be adapted to perform the partitioned grouping. The results of these algorithms with automatic clustering will be evaluated through internal and external validation indexes. (AU)