A wide variety of applications like patient monitoring, rehabilitation sensing, sports and senior surveillance require a considerable amount of knowledge in recognizing physical activities of a person captured using sensors. The goal of human activity recognition is to identify human activities from a collection of observations based on the behavior of subjects and the surrounding circumstances. Movement is examined in psychology, biomechanics, artificial intelligence and neuroscience. To be specific, the availability of pervasive devices and the low cost to record movements with machine learning (ML) techniques for the automatic and quantitative analysis of movement have resulted in the growth of systems for rehabilitation monitoring, user authentication and medical diagnosis. The self-regulated detection of human activities from time-series smartphone sensor datasets is a growing study area in intelligent and smart healthcare. Deep learning (DL) techniques have shown enhancements compared to conventional ML methods in many fields, which include human activity recognition (HAR). This paper presents an improved wolf swarm optimization with deep learning based movement analysis and self-regulated human activity recognition (IWSODL-MAHAR) technique. The IWSODL-MAHAR method aimed to recognize various kinds of human activities. Since high dimensionality poses a major issue in HAR, the IWSO algorithm is applied as a dimensionality reduction technique. In addition, the IWSODL-MAHAR technique uses a hybrid DL model for activity recognition. To further improve the recognition performance, a Nadam optimizer is applied as a hyperparameter tuning technique. The experimental evaluation of the IWSODL-MAHAR approach is assessed on benchmark activity recognition data. The experimental outcomes outlined the supremacy of the IWSODL-MAHAR algorithm compared to recent models.

Diagnosing and treating newborn seizures accurately and promptly is crucial for providing the best possible care for these patients. For the purpose of intelligently identifying newborn seizures, this work introduced a unique method that uses spectral and spatial graph neural networks (SSGNNs) optimized with the Aquila algorithm. Using electroencephalogram (EEG) recordings, the suggested methodology takes advantage of the complex spatial and spectral characteristics of infant brain activity. Spatial and spectral GNNs were used to extract significant spatiotemporal patterns suggestive of seizure episodes by organizing the brain activity data as a graph, with nodes representing various brain regions and edges signifying functional relationships. By combining spectral and spatial data, the depiction of newborn brain dynamics was improved and made it possible to distinguish between seizure and non-seizure phases with greater accuracy. Moreover, the introduction of the Aquila algorithm improved the GNNs' performance in seizure identification tasks by streamlining the training process. A large dataset of EEG recordings from newborns with and without seizures was used to assess the effectiveness of the suggested method. Higher accuracy, sensitivity, and specificity in seizure detection were achieved in the experimental results, which showed greater performance when compared to conventional methods. This work offered an automated, data-driven method for identifying newborn seizures, which is a major development in the treatment of newborns. By combining spectral and spatial GNNs and optimizing the results using the Aquila method, it is possible to enhance seizure detection accuracy and potentially prevent neurological consequences in affected children by intervening early. This method has the potential to completely change the way neonatal care is provided by giving medical professionals a strong tool for accurate and prompt seizure monitoring in neonatal intensive care units (NICU).