Dissertation defense (December 17, 2020): Luciana Escobar Gonçalves Vignoli
Student: Luciana Escobar Gonçalves Vignoli
Title: Comparative Analysis of Methods for Events Detecting in Time Series
Advisors: Laura silva de Assis (advisor) e Eduardo Soares Ogasawara (co-advisor)
Committee: Laura Silva de Assis (president), Eduardo Soares Ogasawara (CEFET/RJ), Rafaelli de Carvalho Coutinho (CEFET/RJ), Fábio André Machado Porto (LNCC)
Day/time: December 17, 2020 / 14h
Room: https://meet.google.com/vtr-zogo-cny
Abstract: Large volumes of data are collected and stored daily, requiring adequate treatment to return valuable information during analysis. These data, when obeying a chronological order of time, consist of time series. Detecting events in these series is an important task in several areas of knowledge, not being restricted to Information Technology. Events can represent an abnormality, a change in behavior, or a pattern that is repeated in the series. Several methods present in the literature seek to identify a single type of event, however, a smaller amount addresses this detection
in a more generalized way. This dissertation proposes a comparative analysis of different methods for detecting events in time series, involving the identification of anomalies and change points. This comparison is performed through statistical methods based on the moving average, decomposition process, and neighborhood-based techniques. Computational experiments were performed with synthetic and real data involving datasets from different areas of knowledge such as water quality monitoring, data traffic from Yahoo, and oil exploration processes. The results obtained were promising and showed that each data set has its particularity, and it is very important to analyze which method is best suited to a specific set, where a good choice can result in up to 0.99 precision in detecting real data.
in a more generalized way. This dissertation proposes a comparative analysis of different methods for detecting events in time series, involving the identification of anomalies and change points. This comparison is performed through statistical methods based on the moving average, decomposition process, and neighborhood-based techniques. Computational experiments were performed with synthetic and real data involving datasets from different areas of knowledge such as water quality monitoring, data traffic from Yahoo, and oil exploration processes. The results obtained were promising and showed that each data set has its particularity, and it is very important to analyze which method is best suited to a specific set, where a good choice can result in up to 0.99 precision in detecting real data.
Dissertation 
