An evolutionary data-conscious artificial immune recognition system
Poh, Chueh Loo
Kitney, Richard I.
Date of Issue2013
Annual conference on Genetic and evolutionary computation conference (15th : 2013)
School of Chemical and Biomedical Engineering
Artificial Immune Recognition System (AIRS) algorithm offers a promising methodology for data classification. It is an immune-inspired supervised learning algorithm that works efficiently and has shown comparable performance with respect to other classifier algorithms. For this reason, it has received escalating interests in recent years. However, the full potential of the algorithm was yet unleashed. We proposed a novel algorithm called the evolutionary data-conscious AIRS (EDC-AIRS) algorithm that accentuates and capitalizes on 3 additional immune mechanisms observed from the natural immune system. These mechanisms are associated to the phenomena exhibited by the antibodies in response to the concentration, location and type of foreign antigens. Bio-mimicking these observations empower EDC-AIRS algorithm with the ability to robustly adapt to the different density, distribution and characteristics exhibited by each data class. This provides competitive advantages for the algorithm to better characterize and learn the underlying pattern of the data. Experiments on four widely used benchmarking datasets demonstrated promising results -- outperforming several state-of-the-art classification algorithms evaluated. This signifies the importance of integrating these immune mechanisms as part of the learning process.