<![CDATA[Human activity recognition (HAR) has become an important field of study because of its wide range of applications in healthcare, security, and smart living systems. Radio frequency (RF)-based HAR offers a non-invasive and privacy-preserving alternative to traditional vision-based systems. This study proposes a hybrid deep learning model combining long short-term memory (LSTM) networks with Random Forest classifiers for RF-based HAR, aiming to improve recognition accuracy across diverse environments. The model was evaluated using channel state information (CSI) and received signal strength indicator (RSSI) features under line-of-sight (LOS) and non-line-of-sight (NLOS) conditions. synthetic minority over-sampling technique (SMOTE) was integrated to balance the dataset and K-fold cross-validation was employed to assess robustness. The dataset included data from 8 subjects performing 10 different activities. The model achieved high classification accuracy, with 99.40% in Environment 1 (LOS), 97.58% in Environment 2 (LOS), and 98.30% in Environment 3 (NLOS), demonstrating the models adaptability and effectiveness. The results highlight the potential of the hybrid LSTM with random forest approach for scalable and reliable RF-based HAR systems that can be integrated into real-world Internet-of-Things (IoT) applications.]]>
Copyrights © 2025
