Deep learning techniques have been widely applied to Human Activity Recognition (HAR), but a specific challenge appears when HAR systems are trained and tested with different subjects. This paper describes and evaluates several techniques based on deep learning algorithms for adapting and selecting the training data used to generate a HAR system using accelerometer recordings. This paper proposes two alternatives: autoencoders and Generative Adversarial Networks (GANs). Both alternatives are based on deep neural networks including convolutional layers for feature extraction and fully-connected layers for classification. Fast Fourier Transform (FFT) is used as a transformation of acceleration data to provide an appropriate input data to the deep neural network. This study has used acceleration recordings from hand, chest and ankle sensors included in the Physical Activity Monitoring Data Set (PAMAP2) dataset. This is a public dataset including recordings from nine subjects while performing 12 activities such as walking, running, sitting, ascending stairs, or ironing. The evaluation has been performed using a Leave-One-Subject-Out (LOSO) cross-validation: all recordings from a subject are used as testing subset and recordings from the rest of the subjects are used as training subset. The obtained results suggest that strategies using autoencoders to adapt training data to testing data improve some users’ performance. Moreover, training data selection algorithms with autoencoders provide significant improvements. The GAN approach, using the generator or discriminator module, also provides improvement in selection experiments.
Adaptation and Selection Techniques Based on Deep Learning for Human Activity Recognition Using Inertial Sensors
