Detecting Dynamical Changes in Data Streams
Fausto Guzzo da Costa1; Rodrigo Fernandes de Mello1
1 Universidade de São Paulo
Resumo
Nature processes are typically nonstationary, many of them exhibit chaos. In some situations, the changes in their behavior represent the most interesting event. A natural way of analyzing such data is by the study of phase space trajectories over time. In this study, we employ concepts from Data Streams, a subarea of Machine Learning, for detecting such dynamical changes. The proposed method is capable of dealing with data continuously produced at high rates. Experiments with synthetic confirm our approach detects dynamical changes on nonlinear and nonstationary data with noise added.
Palavras-chave: Nonlinear Dynamics and Complex Systems; Chaos and Global Nonlinear Dynamics; Time Series Analysis; Data Streams