Artificial Immune Systems for Diagnostic Classification Problems

Artificial Immune Systems (AISs), a class of artificial intelligence algorithms, have been an area of growing research and development in recent years. AISs, along with other wellknown algorithms such as neural nets or particle swarm optimization, are biologically inspired, with AISs in particular designed to exhibit many of the behaviors of biological immune systems. In this paper, we explore the application of AISs to classification problems, particularly in the context of diagnostics, where the goal is generally to classify data into “nominal” or “error” classes. In particular, we present a formal definition of feature space as a multi-dimensional space constructed by a set of real-valued functions, define the process of feature selection, and explain and demonstrate its importance. We provide an overview of an AIS-based program developed for the International Diagnostic Competition, with particular focus on feature selection and AIS detector generation. Finally, we present experimental results, conclusions, and areas for future research.