Institute of Fundamental Technological Research

Artificial immune systems (AIS) have become popular among researchers and have been applied to a variety of tasks. Developing supervised learning algorithms based on metaphors from the immune system is still an area in which there is much to explore. In this paper a novel supervised immune algorithm based on clonal selection framework is proposed. It evolves a population of linear classifiers used to construct a set of classification rules. Aggregating strategies, such as bagging and boosting, are shown to work w ell with the proposed algorithm as the base classifier.

artificial immune systems, clonal selection, linear classifiers, bagging, boosting

During the last decade artificial immune systems have drawn much of the researchers’ attention. All the work that has been done allowed to develop many interesting algorithms which come in useful when solving engineering problems such as data mining and analysis, anomaly detection and many others. Being constantly developed and improved, the algorithms based on immune metaphors have some limitations, though. In this paper we elaborate on the concept of a novel artificial immune algorithm by considering the possibility of combining the clonal selection principle and the well known K-means algorithm. This novel approach and a new way of performing suppression (based on the usefulness of the evolving lymphocytes) in clonal selection result in a very effective and stable immune algorithm for both unsupervised and supervised learning. Further improvements to the cluster analysis by means of the proposed algorithm, immune K-means, are introduced. Different methods for clusters construction are compared, together with multi-point cluster validity index and a novel strategy based on minimal spanning tree (mst) and a analysis of the midpoints of the edges of the (mst). Interesting and useful improvements of the proposed approach by means of negative selection algorithms are proposed and discussed.

Artificial immune systems, Clonal selection, Negative selection, Clustering, Data analysis

The paper presents a new algorithm for feature selection and classification. The algorithm is based on an immune metaphor, and combines both negative and clonal selection mechanisms characteristic for B- and T-lymphocytes. The main goal of the algorithm is to select the best subset of features for classification. Two level evolution is used in the proposed system for detectors creation and feature selection. Subpopulations of evolving detectors (T-lymphocytes) are able to discover subsets of features well suited for classification. The subpopulations cooperate during evolution by means of a novel suppression mechanism which is compared to the traditional suppression mechanism. The proposed suppression method proved to be superior to the traditional suppression in both recognition performance and its ability to select the proper number of subpopulations dynamically. Some results in the task of ECG signals classification are presented. The results for binary and real coded T-lymphocytes are compared and discussed.

Artificial immune system, Feature selection, ECG signals classification, Negative selection, Clonal selection, Evolutionary feature selection, Evolutionary algorithms, Immune metaphors, Hybrid immune algorithm