Revolutionizing Human Activity Recognition in Healthcare: Harnessing Red Deer for Feature Selection and Focal Loss-Based MLP for Classification

Internet of Things (IoT) and Artificial Intelligence (AI) advancements have notably improved Human Activity Recognition (HAR) for healthcare, using sensor-based wearables to monitor daily and health activities. However, accuracy challenges arise in detecting specific movements like walking upstairs and walking downstairs due to limited training data and complex behaviours. Traditional HAR methods depend on large, accurately labelled datasets, but practical implementations often suffer from reduced accuracy when test subject data is scarce. Additionally, identifying key features for precise classification remains a critical task. In response, we present the Advanced Human Activity Recognition (AHAR) methodology. AHAR employs a bio-inspired feature selection technique to create an enriched feature-selected dataset. This dataset is then subjected to classification using the Multi-Layered Perceptron (MLP) model with focal loss, which helps focus on weakly labelled data as well. When evaluated on the UCI-HAR dataset, our proposed methodology demonstrates significantly improved recognition accuracy by 2% compared to prevailing state-of-the-art techniques, thus providing a robust and reliant HAR system.