Menu Close

Classification/Prediction

Classification/Prediction

 In this topic we investigate new methods for multidimensional sequence classification. We propose a benchmark and new classification techniques based on time series shapelets, that deal with sequence and are able to automatically combine multiple dimensions.

Carlos Andres Ferrero; Luis Otavio Alvares; Willian Zalewski; Vania Bogorny. 

In: ACMSAC, 2018. Proceedings of the 33 Symposium on Applied Computing, Pau, France, April 9-13, 2018

Movelets finds the best multidimensional subsequences that discriminate the class. It is a generic method that works for any sequencial data classification problem, being able to deals the dimensions space, time, semantics and others. Movelets uses local features (subsequences) and global features extracted from the entire sequence.

FERRERO, Carlos Andres, PETRY, Lucas May, ALVARES, Luis Otavio, SILVA, Camila Leite, ZALEWSKI, Willian, BOGORNY, Vania.

In: Data Mining and Knowledge Discovery (2019).   

MasterMovelets finds the most discriminant subsequences and automatically chooses the best dimension combination for each subsequence. Subsequences can be heterogenous, so a dimension that is discriminant for one class or one subsequence, may not be for another. MasterMovelets is robust for multiple aspect mobility data, biological and health data.

SILVA, Camila Leite, PETRY, Lucas May, BOGORNY, Vania.

In: international Journal of Geographical Information Science (2019)

This works presents a benchmarck for mobility data classification and proposes the method Pivot, which uses as input only the best movelets of size one and uses the neighobour points to generate movelets of different size. This method optimizes the processing time of both movelets and mastermovelets and keeps the accuracy.

SILVA, Camila Leite, PETRY, Lucas May, BOGORNY, Vania.

In: 2019 Brazilian Conference on Intelligent Systems (BRACIS). 2019. IEEE. ed 8, p. 788-794.

Compares and sumarizes existing works for trajectory classification.