An immune approach to classifying the high-dimensional datasets

This paper presents an immune-based approach to problem of binary classification and novelty detection in high-dimensional datasets. It is inspired by the negative selection mechanism, which discriminates between self and nonself elements using only partial information. Our approach incorporates two types of detectors: binary and real-valued. Relatively short binary receptors are used for primary detection, while the real valued detectors are used to resolve eventual doubts. Such a hybrid solution is much more economical in comparison with “pure” approaches. The binary detectors are more faster than real-valued ones, what allows minimize computationally and timely complex operations on real values. Additionally, regardless of type of encoding, the process of sample’s censoring is conducted with relatively small part of its attributes.