Dissertation defense (March 05, 2021): Antonio Jose de Castro Filho

Student: Antonio Jose de Castro Filho

Title: Mining of Space and Time Constrained Sequences

Advisors: Rafaelli Coutinho (advisor) and Eduardo Ogasawara (co-advisor)

Committee: Rafaelli Coutinho (president), Eduardo Ogasawara (CEFET/RJ), Jorge Soares (CEFET/RJ), Esther Pacitti (INRIA)

Day/Time: March 05, 2021 / 10h.

Room: meet.google.com/eck-qenz-vap

Abstract: Spatio-temporal patterns bring knowledge about time and position where they are present. Finding them is an important task for different domains. However, not all patterns are frequent over an entire dataset, they can occur constrained in space and time. Mining these patterns have as objective to discover the time range, and the set of spatial positions in which event sequences are frequent. This work proposes Generalized Spatial-Time Sequence Miner (G-STSM) algorithm as a solution for the discovery of frequent sequences that are constrained in space and time, bringing the formalization of the problem, definitions, proofs, and algorithms. As far as is known, after searching the related literature, G-STSM is the first approach able to find such sequences working with one dimension of time and three dimensions of space. G-STSM has been compared with an intuitive approach that searches for sequences of frequent events with very low support and groups its occurrences to find patterns constrained in space and time using known algorithms. A set of real-world space-time seismic dataset was chosen to compare both approaches using classification metrics and resource usage records. As a result, G-STSM presented better computational performance with similar quality and it proved to be an efficient data mining tool for finding tight space-time sequences.