Human Activity Recognition (HAR) has become increasingly significant in domains such as healthcare, surveillance, human-computer interaction, and smart environments. This study proposes an Artificial Intelligence and Machine Learning (AIML)-driven framework for recognizing human activities using a hybrid MobileNet-LSTM model optimized by the Wildebeest Herd Optimization Algorithm (WHOA). Leveraging multimodal datasets that include RGB, depth, and skeletal joint information, the system applies rigorous preprocessing steps such as frame segmentation, temporal sampling, optical flow extraction, and data augmentation to ensure data reliability and model robustness. MobileNet is employed for efficient spatial feature extraction, while LSTM captures the temporal dependencies within action sequences. WHOA is used to fine-tune hyperparameters, improving the overall model accuracy and performance. Experimental results demonstrate that the framework achieves high recognition rates while maintaining low computational complexity, making it suitable for real-time applications on resource-constrained devices.

Human Activity Recognition, AIML, MobileNet, LSTM, WHOA, Deep Learning, Smart Monitoring, Real-Time Systems