Título: Streaming Data Classification with the K-associated Graph
Autores: Bertini Jr., João R.; Lopes, Alneu; Zhao, Liang
Resumo: Graph-based methods consist in a powerful form for data representation and abstraction, such approach has been each time more considered to solve machine learning related problems. This work presents a graph-based algorithm suitable for streaming data classification problems with non-stationary distribution. The proposed algorithm is an extension of a static algorithm which relies on the following concepts, a graph structure called optimal K-associated graph, that represents the training set as a sparse graph separated into components; and the purity measure for each component, which uses the graph structure to determine local data mixture level in relation to their classes. Here we address classification problems that exhibit modification on the underlying data distribution. This problem qualifies concept drift and worsens any static classifier performance. Hence, in order to maintain accuracy performance, it is necessary for the classifier to keep learning during application phase, for example, by implementing incremental learning. Experimental results, suggest that our algorithm presents advantages over the tested algorithms on streaming classification tasks.
Palavras-chave: Graph-based Learning; Streaming Data Classification; Incremental Learning
Código DOI: 10.21528/CBIC2011-27.2
Artigo em pdf: st_27.2.pdf
Arquivo BibTex: st_27.2.bib