SAINET com raio adaptativo: estimativa do raio mínimo e modificação no cálculo da densidade

Abstract

There are several algorithms based on the most different approaches to perform data clustering and data classification in literature. Among those based on density, some are designed to generate prototypes in order to represent the dataset reducing the volume of data managed. One of the first of these immune-inspired is the Artificial Immune Network (aiNet), which is an artificial immune network designed to perform data clustering. From aiNet, many other algorithms have been proposed to be applied to other kind of problems. A supervised version of aiNet was porposed to perform data classification, called SAINET (supervised AINET). However, aiNet did not present satisfactory performance when applied to certain data distributions and another network was presented: the Artifical Radius Immune Algorith (ARIA) aiming to achieve better results where aiNet has not seem to be efficient. Based on that, a modification on SAINET original algorithm was proposed by replacing aiNet impementation by ARIA on building the network population, being though a SAINET wiht adaptative radius. Literature presents several studies about ARIA’s efficiency regarding density preservation. Nevertheless, not much has been said about how to set its parameters that are not adjusted by the algorithm. The present work proposes a possibility to adjust ARIA’s antibodies minimum radius, a parameter that is set by the user which is extremely important to network’s performance and is not clear how to do. Further, we inted to verify whether the modification proposed to ARIA in order to preserve density affects SAINET classification performance.