Predictions and patterns identification

The extraction of knowledge from the data is an extremely important activity and demanded by several organizations existent in the business, governmental and scientific axis. A subset of important problems encompasses the study of time and space-time series. The time series corresponds to a sequence of observations, while the space-time series have a position associated with that sequence. These observations can be univariate, multivariate and have different domains and periodicity, bringing a heterogeneous character to the data. Time-series searches commonly rely on the exploration of representation, indexing, prediction, classification, and pattern identification (motifs) methods on these data. There are a large number of models for predicting and classifying time series. Each of them has different properties and complexities. Some of them represent state of the art in machine learning methods. In addition, time series management involves the analysis of different properties such as seasonality, stationarity, and homoscedasticity. These properties allow the possibility of exploring different transformation methods, filtering and adjustment models to obtain satisfactory predictions. Finally, the identification of frequent and previously unknown patterns in time series, also called motifs, has also been well researched. However, in the context of space-time series, there is still a gap that is not well explored and at the same time rich in challenges. The phenomena described by such models bring challenges both in data management and in the conception of methods for knowledge extraction.