Graph models for machine learning: K-associated graphs and Attribute-based Decision Graphs

Graph-based methods consist of a powerful form of data representation and abstraction. Among other features, they allow uncovering topological relationships, visualizing structures, representing groups of data with different formats, and providing alternative measures to characterize the data. However, the majority of traditional graph-based methods, in general, presents a high computational complexity order, which limits the scope of application of these methods. In this context, alternative approaches based on graphs, such as K-associated graphs and Attribute-based Decision Graphs, present low computational complexity and, at the same time, have the advantages of graph-based learning. Taking into account the advantages of using graphs to represent data and the success of this kind of approach, this project aims to extend the models of K-associated graphs and decision graphs to solve data mining problems not yet treated by them, such as: data clustering, novelty detection in data stream, active learning applied to data stream, among others.