Advanced search
Start date
Betweenand

Providing theoretical guarantees to the detection of concept drift in data streams

Grant number: 17/16548-6
Support type:Scholarships abroad - Research
Effective date (Start): August 01, 2018
Effective date (End): January 31, 2019
Field of knowledge:Physical Sciences and Mathematics - Computer Science
Principal Investigator:Rodrigo Fernandes de Mello
Grantee:Rodrigo Fernandes de Mello
Host: Albert Bifet
Home Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Local de pesquisa : ParisTech, France  
Associated research grant:13/07375-0 - CeMEAI - Center for Mathematical Sciences Applied to Industry, AP.CEPID

Abstract

With the objective of modeling data stream changes, several researchers have been designing new approaches to detect concept drifts. A concept is characterized by a sequence of observations produced by a same generating process. Researchers are interested in detecting concept drift in order to support specialists to make decisions on the phenomena that generated such streams. Currently, there are two main research areas devoted to the concept drift detection: the first is based on Supervised learning, while the second relies on unsupervised approaches. Both lack in terms of providing theoretical guarantees while detecting drifts, once the first relaxes the assumption of data independency, required by the Empirical Risk Minimization Principle defined in the context of the Statistical Learning Theory, and the second fails due to no theoretical framework ensures learning, therefore detections are usually caused by the algorithm parametrization and not due to data changes. In order to tackle such drawbacks, this research project aims at formulating a theoretical framework to ensure that detections of concept drift in data streams are due to modifications occurred in in the observations collected along time and not by chance nor parametrization. Furthermore, we will design and implement an algorithm to detect concept drift under such theoretical guarantees. Experiments will be performed using transitions among data stream concepts produced by different synthetic generating processes, as well as by real-world streams.

Scientific publications
(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)
DE MELLO, RODRIGO F.; VAZ, YULE; GROSSI, CARLOS H.; BIFET, ALBERT. On learning guarantees to unsupervised concept drift detection on data streams. EXPERT SYSTEMS WITH APPLICATIONS, v. 117, p. 90-102, MAR 1 2019. Web of Science Citations: 0.
DE MELLO, RODRIGO F.; RIOS, RICARDO A.; PAGLIOSA, PAULO A.; LOPES, CAIO S. Concept drift detection on social network data using cross-recurrence quantification analysis. Chaos, v. 28, n. 8 AUG 2018. Web of Science Citations: 0.

Please report errors in scientific publications list by writing to: cdi@fapesp.br.