Alohali and colleagues introduced a new system called AHARDP-DLSO (Advanced Smart Human Activity Recognition for Disabled People Using Deep Learning with Snake Optimizer). The goal of the Human Activity Recognition (HAR) approach is to better recognize daily activities in people with disabilities, which is crucial for independence and healthcare monitoring. Unlike prior studies that primarily rely on conventional machine learning or deep learning models without adaptive optimization, AHARDP-DLSO combines three main steps: Min-max normalization for consistent input scaling. Deep Belief Networks (DBN) to classify different activities by learning patterns from sensor data. Snake Optimizer Algorithm (SOA) to fine-tune model parameters and improve accuracy. The system was validated on the WISDM dataset, which includes six activity classes (walking, jogging, sitting, standing, upstairs, downstairs). They tested the method on the WISDM dataset, which includes six types of activity (walking, jogging, sitting, standing, going upstairs, and going downstairs). AHARDP-DLSO reached an accuracy of 95.81%, beating out other strong models like CNN-LSTM (95.25%), At-CapNet (92.33%), and EfficientNet B0 (89.11%). Importantly, it was not only more accurate but also more efficient, with a computation time of just 2.12 seconds, much lower than other approaches. By effectively leveraging AI with optimization techniques, AHARDP-DLSO improves both precision and computational efficiency, addressing gaps in adaptability, robustness, and real-world deployment. That said, the system was only tested on one dataset, so it’s not clear how well it would hold up in more varied, real-world settings. Things like sensor placement, noise, or differences across users could still affect performance. Testing across multiple datasets and exploring deployment on wearables or edge devices would be useful next steps to show its reliability outside the lab.