A specialized lazy learner for time series forecasting

TitleA specialized lazy learner for time series forecasting
Publication TypeConference Paper
Year of Publication2017
AuthorsMartínez, Francisco, Frías M.P., Charte Francisco, and Rivera-Rivas A.J.
Conference Name17th International Conference on Computational and Mathematical Methods in Science and Engineering, CMMSE 2017
Pagination1397–1403
Date Published7
Conference LocationCosta Ballena, Rota, Cáadiz (Spain)
ISBN Number978-84-617-8694-7
Abstract

In a time series context the nearest neighbour algorithm looks for the historical observations most similar to the latest observations of the time series. However, some nearest neighbours can be misleading. In this paper we propose that, if prior information about the structure of the time series is known, the search space of possible neighbours can be narrowed so that some possibly misleading neighbours are avoided. This way a more effective forecasting method can be obtained.

Notes

TIN2015-68854-R