Title: Multi-classification of human body motions from acceleration curves
Authors: M Carmen Aguilera-Morillo - Universidad Carlos III de Madrid (Spain) [presenting]
Ana Maria Aguilera - University of Granada (Spain)
Abstract: The aim is to classify a set of functional data according to a multinomial response variable. Exactly, a set of accelerations curves measured during the realization of three different body motions (walking, walking upstairs and walking downstairs) have been considered. In order to solve this multi-classification problem, methodology based on functional linear discriminant analysis (FLDA) of the multinomial response variable on a set of functional PLS components of the acceleration curves is proposed. With the aim of improving both, the classification of new samples curves and the estimation of the discriminant functions, two penalized versions of functional PLS regression are combined with FLDA. The results obtained from this dataset highlight the need to use a penalization term, in which case a correct classification rate greater than the 80$\%$ has been achieved on the test sample.