Hybrid deep neural network models for boosting Human Activity Recognition using IoT wearables
S. Sowmiya, D MENAKA · Research Square · 2022
Abstract Human Activity Recognition is a key element for many immense applications in human life. With the advances in sensor technologies and utilizing the IoT, HAR has a wide area of research with the help of deep learning algorithms. The advanced deep learning paradigm provides end-to-end learning models from unstructured, complex data. IoT wearables and smartphones are now widely used embedded with mobile apps for telemedicine, e-health monitoring, sports monitoring, AAL, biometrics, smart homes etc. This paper presents hybrid neural networks model implemented with Bidirectional GRU, Bidirectional LSTM and CNN. The algorithm was tested using three activity recognition datasets WISDM ,USCHAD and MHEALTH. The hybrid model provides improved accuracy over the other activity recognition techniques.