Evolving clustering, classification and regression with TEDA

Abstract

In th i s art i cle the novel cluster i ng and regress i on methods TEDACluster and TEDAPredict methods are descr i bed additionally to recently proposed evolv i ng class i fier TEDAClass. The algor i thms for class i fication, cluster i ng and regression are based on the recently proposed AnYa type fuzzy rule based system. The novel methods use the recently proposed TEDA framework capable of recursive process i ng of large amounts of data. The framework i s capable of computationally cheap exact update of data per sample, and can be used for training 'from scratch'. All three algor i thms are evolv i ng that is they are capable of chang i ng i ts own structure dur i ng the update stage, wh i ch allows to follow the changes with i n the model pattern.

Metodologia Incremental para Agrupamento em Fluxos Contínuos de Dados

“…Devidoà rapidez e facilidade de implementação, extensões do TEDA têm sido propostas para lidar com problemas de classificação [9] , regressão [8] e agrupamento [6] em fluxos contínuos de dados. Uma característica importante deste modeloé que a tipicidade se assemelha (embora seja diferente)à uma função massa de probabilidade [4], sendoútil para aplicações onde se faz necessário o uso de soft clustering ou modelar a distribuição dos dados quando estaé desconhecida a priori.…”

Developing Communication Strategy for Multi-Agent Systems with Incremental Fuzzy Model

“…In [20] a simple and fast method is proposed that uses gradient decent to www.ijacsa.thesai.org tune the structure and parameters of a fuzzy classifier. D. Kangin et al in [21] and [22] have introduced a group of incremental methods called TEDA that can be used for clustering, regression and classification. Incremental methods are also employed to find a policy for infinite-horizon Dec-POMDP.…”

Fast feedforward non-parametric deep learning network with automatic feature extraction

“…The classification results are tabulated in Table I compared with the results of the previously published methods [16]- [18]. The results are visualized in Fig.…”