Multi-label classification in data-streams

Abstract

Data Streams are unlimited data sequences, continuously generated, non-stationary, and in many cases arriving in high speed. Novelty Detection is a classification task that verifies if an example, or a set of examples, differs significantly from the previously seen examples. This is an important task for Data Stream Classification, since new concepts may appear, disappear, or evolve over time (Concept Evolution). MINAS is a Novelty Detection algorithm which considers the classification task as multi-class problem, and is able to detect the previously mentioned changes in data. Most applications in Data Stream Classification are multi-label. However, few works in the literature focused on multi-label classification. Therefore, the main objective of this project is to adapt the MINAS algorithm for multi-label classification, considering Concept Drifts and Concept Evolution. Besides, techniques that recognize label dependencies will be studied to improve the classifier performance. The proposed method will be compared with those existing in the literature and evaluated through specific measures for multi-label classification problems and Concept Evolution detection. (AU)