Evaluating Linear Models as a Baseline for Time Series Imputation

A aluna Rebecca Salles apresentou o trabalho intitulado “Evaluating Linear Models as a Baseline for Time Series Imputation” no XXX Simpósio Brasileiro de Banco de Dados (SBBD 2015). O trabalho está no escopo de sua Iniciação Científica orientada pelo Prof. Eduardo Ogasawara. Colaboraram no artigo os professores Eduardo Bezerra e Jorge Soares.

2015-10-14 14.17.35

2015-10-14 14.17.59

Abstract: Time series prediction has been gaining attention of many researchers throughout the world for its increasing importance to preparation, planning and decision-making activities in many areas of study in science, business and government. Many data come from different sources and some of them, such as sensors, are not resilient to failures. A particular problem that occurs in these cases is the absence of data in some parts of the time series. Addressing this lack of data becomes important to enable the development of prediction models. Although there are many machine learning methods (MLM) that may be used to fill such data, there is an absence of systematically benchmarking established linear baseline methods for performance comparison. In this paper we explore linear models as baseline for time series imputation (TSI). Our results show the importance of exploring different linear approaches for TSI to encourage researchers to improve their choices for a suitable MLM for solving such problem.

Artigo completo

Comments are closed.