Full Program »
Roll Padding and Wavenet For Multivariate Time Series In Human Activity Recognition
Padding is a process used for the border treatment of data before the convolution operation in Convolutional Neural Networks. This study proposes a new type of padding designated by roll padding, which is conceived for multivariate time series analysis when using convolutional layers. The Human Activity Recognition raw time distributed dataset is used to train, test and compare four Deep Learning architectures: Long Short-Term Memory, Convolutional Neural Networks with and without roll padding, and WaveNet with roll padding. Two main findings are obtained: on the one hand, the inclusion of roll padding improves the accuracy of the basic standard Convolutional Neural Network and, on the other hand, WaveNet extended with roll padding provides the best performance result.