Anomaly Detection Using Neighborhood Negative Selection

Abstract

Abstract Negative Selection Algorithms (NSAs) have been widely used in anomaly detection. As the security issue becomes more complex, more and more anomaly detection schemes involve high-dimension data. NSAs however perform poorly on effectiveness and efficiency when dealing with high-dimension data. To address these issues, we propose a Neighborhood Negative Selection (NNS) algorithm in this paper. Instead of a single data point, NNS uses a neighborhood to represent a self sample (or a detector). As a result, the training efficiency is greatly improved. We further introduce a special matching mechanism to limit the negative effect of the dimensionality of a shape space and improve the detecting performance in high dimensions. The experimental results show that NNS can provide a more accurate and stable detection performance. Meanwhile, both theoretical analysis and experimental results show that NNS further improves the training efficiency.