Detectores de mudança de conceito por meio do mapeamento espacial do fluxo de dados usando quadtree

Abstract

Online learning is a complex task, especially when the data stream changes its distribution over time. It has been a challenge to monitor and detect these changes to preserve the learning algorithm performance. This work presents two novels drift detection methods built from a different perspective of other preexisting detectors from literature.It analyzes the space occupied by the data, assuming that it would be immutable unless changes in this space occur among data of different classes. Data are mapped into a quadtree-based memory structure that provides knowledge about which class (label) is dominant in a given region of the feature space. The proposed method QT detects a drift by checking whether data assigned to a given class occupy spaces considered relevant to the other class. The QTS, on the other hand, detects a concept drift when it identifies a significant increase in the increment of data in one of the classes. The proposed methods were evaluated on binary classification benchmark problems. Results show that our methods were competitive with well-known drift detectors from literature.