IoT-Based Wrist-Worn Ambient Assisted Living Device with Attention-Enhanced CNN-GRU for Human Activity Recognition

The increasing demand for independent living and personalized healthcare has powered advancements in wearable technology and ambient assisted living (AAL) systems. This paper introduces a wrist-worn AAL device with human activity recognition (HAR) via a hybrid deep learning model and multisensor fusion. The model combined Convolutional Neural Networks (CNN) for spatial feature extraction, Gated Recurrent Units (GRU) for processing time-series data, and an Attention Mechanism (AM) to highlight critical features across the time dimension. Integrating accelerometer-gyroscope data with heart rate and blood oxygen data significantly improved HAR's classification results, achieving an accuracy and recall of 92.59%, precision of 93.78%, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$f1$</tex>-score of 92.47%. The prototype also demonstrated effective internet of things (IoT) integration, achieving real-time HAR and vital signs data transmission over the internet and removing the dependency on wired connections to external processors. These findings highlight the potential of compact wearable IoT devices and advanced deep learning models with multi-sensor fusion in addressing the growing need for independent living and personalized healthcare solutions.