A systematic literature review on human activity recognition using smart devices: advances, challenges, and future directions

With improvement in AI and deep learning, Human Activity Recognition (HAR) has started to play an important role in many real-world applications. These range from healthcare and elderly care to security to enhancing user experience in smart environments. Smart devices with their ubiquity and multi-modal sensors, provide critical data that can be used to improve the quality of life and efficiency of services. Given the rapid technological advancements and recent developments in the field, it is essential to examine the current state of Human Activity Recognition (HAR), highlighting its strengths as well as the challenges it continues to face. This study presents a systematic review of the literature (SLR) of HAR that is conducted through a detailed study of articles published from 2021 to 2024 using open data sets. This study explores major databases i.e., Springer, IEEE Xplore, Science Direct, Taylor & Francis, and ACM to access the recent advancements in HAR that are facilitated by smartphones and smartwatches. Our SLR looks at the most commonly used datasets i.e. WISDM, PAMAP2, UCI-HAR, Opportunity, and MHealth. Moreover, this SLR compares the performance of commonly used deep learning techniques i.e. Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM), Transformers and Ensemble models on these datasets. The objectives of this study are threefold i.e. to identify recent studies that employ deep learning techniques for Human Activity Recognition (HAR) tasks, to review publicly available open datasets commonly used in HAR research during 2021 to 2024, and to identify common challenges and limitations in current HAR research while proposing potential future research directions.