Utilisation of Exponential-Based Resource Allocation and Competition in Artificial Immune Recognition System

There has been a rapid growth in using Artificial Immune Systems for applications in data mining and computational intelligence recently. There are extensive computational aspects with the natural immune system. Several algorithms have been developed by exploiting these computational capabilities for a wide range of applications. Artificial Immune Recognition System is one of the several immune inspired algorithms that can be used to perform classification, a data mining task. The results achieved by Artificial Immune Recognition Systems have shown the potential of Artificial Immune Systems to perform classification. Artificial Immune Recognition System is a relatively new classifier and has some advantages such as self regularity, parameter stability and data reduction capability. However, the Artificial Immune Recognition System uses a linear resource allocation method. This linearity increases the processing time of generating memory cells from antigens and causes an increase in the training time of the Artificial Immune Recognition System. Another problem with the Artificial Immune Recognition System is related to the resource competition phase which generates premature memory cells and decreases the classification accuracy of system. This thesis proposes new algorithms based on Artificial Immune Recognition System to address the mentioned weaknesses and improve the performance of the Artificial Immune Recognition System. Firstly, exponential-based resource allocation methods are utilized instead of the existing linear resource allocation method. Next, the Real World Tournament Selection method is adapted and incorporated into the resource competition of Artificial Immune Recognition System. The proposed algorithms have been tested on a variety of datasets from the UCI machine learning repository. The experimental results show that utilizing exponential-based resource allocation methods decreases the training time and increases the data reduction capability of Artificial Immune Recognition System. In addition, incorporating an adapted Real World Tournament Selection technique increases the accuracy of the Artificial Immune Recognition System up to 4%. The difference between the performances of the proposed algorithms and Artificial Immune Recognition System are significant in majority of cases.

Preview PDF FSKTM_2011_3_F.pdf Download (247kB) | Preview

Actions (login required)

View Item