Evolving 1D Convolutional Neural Networks for Human Activity Recognition

Human activity recognition is an important research field with a variety of applications in healthcare monitoring, fitness tracking and in user-adaptive systems in smart environments. The problem of human activity recognition can be solved using a 1D convolutional neural network (CNN) trained with accelerometric data. The design of an appropriate CNN architecture for solving a particular problem is not an easy task and usually requires considerable specialized knowledge to setup the network hyperparameters based on experimental evaluation. This article proposes an automated approach for CNN architecture optimization that uses genetic algorithms. The suggested approach for evolution of the architecture of 1D CNN is evaluated on two data sets for accelerometer-based human activity recognition and the results show that the GA based CNN design generates CNN architectures with competitive performance compared to the usage of other manually designed CNN models.