Time Series Data Augmentation and Dropout Roles in Deep Learning Applied to Fall Detection

• Enol García González ^[1] ; José Ramón Villar ^[1] ; Cal, Enrique de la ^[1]
  1. [1] Universidad de Oviedo

     Universidad de Oviedo

     Oviedo, España

     
• Localización: 15th International Conference on Soft Computing Models in Industrial and Environmental Applications (SOCO 2020): Burgos, Spain ; September 2020 / coord. por Álvaro Herrero Cosío , Carlos Cambra Baseca , Daniel Urda Muñoz , Javier Sedano Franco , Héctor Quintián Pardo , Emilio Santiago Corchado Rodríguez , 2021, ISBN 978-3-030-57802-2, págs. 563-570
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • Fall Detection is one of the most interesting and challenging research topics in the world today because of its implications in society and also because the complexity of processing Time Series (TS). Plenty of research has been published in the literature, several of them introducing Deep Learning (DL) Neural Network (NN) as the modelling element. In this study we analyse one of these contributions and address several enhancement using TS data augmentation and dropout. Moreover, the possibility of reducing the NN to make it lighter has been studied. The NN has been implemented using Keras in Python and the experimentation includes an staged fall publicly available data set. Results show the TS data augmentation together with dropout helped in learning a more robust and precise model. Future work includes introducing different types of cross-validation as well as introducing other types of DL models more suitable for TS.