Enhancing Performance in Physical Activity Monitoring: Leveraging Wi-Fi Channel State Information (CSI)

In the domain of physical activity monitoring, the utilization of Wi-Fi Channel State Information (CSI) introduces a novel dimension of data-driven analysis. This study explores the integration of CSI data with Convolutional Neural Network (CNN) architectures to enable the classification and accurate identification of six distinct human activities. A sizable dataset, comprising 6200 images, was meticulously curated and partitioned into a 70:30 ratio for rigorous training and testing of the CNN models. To fine-tune the models, we employed a suite of optimizers, including Adam, RMSprop, and stochastic gradient descent (SGD), each showcasing unique characteristics in optimizing the learning process. A pivotal facet of our approach is the meticulous optimization of hyperparameters, particularly the learning rate, set at an optimal value of 0.001, ensuring the models’ convergence and overall performance in activity recognition. The results of our comprehensive analysis highlight the efficacy of Wi-Fi CSI data in activity classification and provide a comparative evaluation of different optimizers’ performance. Our findings reveal nuanced insights into the choice of optimizers for CNN models in CSI-based applications. Beyond the technical intricacies, this research underscores the potential of Wi-Fi CSI in diverse environments, including healthcare, smart spaces, and ambient-assisted living. By delving into the capabilities and limitations of this technology, we contribute to the broader discourse on data-driven approaches to physical movement monitoring. Our study illuminates the multifaceted applications of Wi-Fi CSI and its role in the evolving landscape of human activity recognition, promising advancements in health monitoring and intelligent environments.