A New Classifier Based on Negative Selection Algorithm

The negative selection algorithm (NSA) is one of the most important artificial immune techniques, which is both an anomaly detection and pattern recognition technique. Recent work, however, demonstrates the successful application of this algorithm in data classification problems. Most of the negative selection-based methods consider explicit and decisive boundary to distinguish between self and non-self spaces. In this paper, an algorithm based on NSA was proposed where a Gaussian Mixture Model (GMM) is fitted onto a normal space to create a flexible boundary between self and non-self spaces and determine the dynamic subset of effective detectors for data classification problems. Since the assignment of optimal values to the effective parameters of artificial immune algorithms plays a crucial role in improving their performance, particle swarm optimization (PSO) was used to optimize the control parameters of the proposed method. Empirical results showed that the use of GMM and the dynamic adjustment of parameters such as the optimum number of Gaussian components, according to the shape of the boundaries, the creation of appropriate number of detectors, and also the automatic adjustment of the thresholds for training and classification, specifically for each class, by using particle swarm optimization algorithm, as well as employing a hybrid objective function, led to a better classification accuracy with fewer detectors on a variety of data sets with completely non-linear boundaries.

کلیدواژه ها:

Classification ، Negative Selection Algorithm ، Gaussian Mixture Model ، Particle Swarm Optimization ، Flexible Boundaries