Semantically Enriching the Detrending Step of Time Series Analysis
Lucelia de SOUZA, Maria Salete Marcon Gomes Vaz, Marcos Sfair Sunye · 2015
In time series analysis, the trend extraction – detrending is considered a relevant step of preprocessing, where occurs the transformation of nonstationary time series in stationary, that is, free of trends. Trends are time series components that need be removed because they can hide other phenomena, causing distortions in further processing. To helping the decision making, by researchers, about how and how often the time series were detrended, the main contribution of this paper is semantically enriching this step, presenting the Detrend Ontology (DO prefix), designed in a modular way, by reuse of ontological resources, which are extended for modeling of statistical methods applied for detrending in the time domain. The ontology was evaluated by experts and ontologists and validated by means of a case study involving real-world photometric time series. It is described its extensibility for methods in time-frequency domain, as well as the association, when applicable, of instances with linked open data from DBpedia semantic knowledge base. As result of this paper, stands out the semantic enrichment of a relevant step of the analysis, contributing to the scientific knowledge generation in several areas that analyze time series.