Multiple Metric Learning for large margin kNN Classification of time series

Cao-Tri Do (1, 2, 3) , Ahlame Douzal-Chouakria (1) , Sylvain Marié (3) , Michèle Rombaut (2)

1 LIG - Laboratoire d'Informatique de Grenoble
2 GIPSA-AGPIG - GIPSA - Architecture, Géométrie, Perception, Images, Gestes
3 SE - Schneider Electric

Time series are complex data objects, they may present noise, varying delays or involve several temporal granularities. To classify time series, promising solutions refer to the combination of multiple basic metrics to compare time series according to several characteristics. This work proposes a new framework to learn a combination of multiple metrics for a robust kNN classifier. By introducing the concept of pairwise space, the combination function is learned in this new space through a "large margin" optimization process. We apply it to compare time series on both their values and behaviors. The efficiency of the learned metric is compared to the major alternative metrics on large public datasets.

Mots clés: Index Terms— Multiple metric learning; Classification; Time series; kNN

Domaines: Machine Learning [stat.ML]; Traitement du signal et de l'image [eess.SP]

Cao-Tri Do, Ahlame Douzal-Chouakria, Sylvain Marié, Michèle Rombaut. Multiple Metric Learning for large margin kNN Classification of time series. EUSIPCO 2015 - 23th European Signal Processing Conference, Aug 2015, Nice, France. ⟨hal-01202045⟩