An Improved Negative Selection Algorithm Based on a T Cell Multilayer Immune Tolerance Mechanism

The negative selection algorithm simulates part of the immune tolerance mechanism, but its simplified detection method limits its application in complex issues. This paper revisits the biological immunology and proposes a detector checking based on a multilayer immune tolerance method(DC-MIT), which focuses on the detection phase rather than the generation phase. Instead of checking directly, the detectors generated from the training set need to be activated through concentration recognition. The testing set needs to be divided into two parts: the activating set and the checking set. First, the activation set is clustered using K-means and divided into different categories. Then, the detectors that can recognize a sufficient number of abnormal samples in the same category are activated and retained through concentration activation to filter out the “bad” and non-functional detectors for this time. Finally, the checking set verifies the activated detectors. A comparison of the NSA algorithm and the classic machine learning algorithms on the Iris and Breast Cancer Wisconsin datasets shows that the method proposed in this paper is superior to other methods, especially in improving precision and false positive issues.