Population-Based Artificial Immune System Clustering Algorithm

Abstract

Artificial immune systems inspired by humoral-mediated immunity use hyper mutation to simulate the way that natural immune systems refine their B cells and antibodies in response to pathogens in a process called affinity maturation. Such hyper mutation is typically performed on individual computational antibodies and B cells, and has been shown to be successful in a variety of machine learning tasks, including supervised and unsupervised learning. This paper proposes a population-based approach to affinity maturation in the problem domain of clustering. Previous work in humoral-mediated immune systems (HAIS), while using concepts of immunoglobulins, antibodies and B cells, has not investigated the use of population-based evolutionary approaches to evolving better antibodies with successively greater affinities to pathogens. The population-based approach described here is a two step algorithm, where the number of clusters is obtained in the first step using HAIS and then in step two a population-based approach is used to further enhance the cluster quality. Convergence in the fitness of populations is achieved through transferring memory cells from one generation to another. The experiments are performed on benchmarked real world datasets and demonstrate the feasibility of the proposed approach. Additional results also show the effectiveness of the crossover operator at the population level. The outcome is an artificial immune system approach to clustering that uses both mutation within antibodies and crossover between members of B cells to achieve effective clustering.