Hybrid Approaches for Time Series Prediction

• Tipo: Artigo em Livro de Atas de Conferência Internacional
• Data: 2018

Autores

Daniel Castro Silva (Autor) FEUP; Xavier Fontes (Autor) FEUP

Ata de Conferência Internacional

Título: Proceedings of the 18th International Conference on Hybrid Intelligent Systems (HIS 2018)

Páginas: 146-155

Indexação

ISI Proceedings

Scopus - 1 Citação

INSPEC

Outras Informações

ID Authenticus: P-00Q-GFD

DOI: 10.1007/978-3-030-14347-3_15

Abstract (EN): The focus of this work is the development of various hybrid prediction models, capable of predicting a given variable in the context of a time series with sporadic external stimuli. As a case study, we use data from a university student parking lot, together with other events and information. Working on top of previous research, we used Gradient Boosting models, Random Forests and Decision Trees in three proposed hybrid approaches: a voting-based combination of models, an approach based on pairs of models working together and a third novel approach, based on social dynamics and trust in human beings, called Evolutionary Directed Graph Ensemble (EDGE). Results show some promise from these methods, in particular from the EDGE approach.

Idioma: Inglês

Tipo (Avaliação Docente): Científica

Nº de páginas: 10