Human activity recognition using marine predators algorithm with deep learning

In the era of smart life, tracking human activities and motion can play a significant role in the advanced modern applications, such as the Internet of things (IoT), Internet of healthcare things (IoHT), smart homes, eldercare, and different health informatics-based applications. Human activity recognition (HAR) has the ability to expose abundant information collected from different devices (i.e., cameras or sensors) that can represent human motion and activities. The recent advances in artificial intelligence methods, including deep learning (DL) and swarm intelligence (SI) optimization algorithms, play a significant role in different applications. In this paper, we integrate the applications of both DL and SI to build a robust HAR system using wearable sensor data. A light feature extraction approach is developed using the residual convolutional network and a recurrent neural network (RCNN-BiGRU). To select the optimal feature set, we develop new feature selection methods based on the marine predator algorithm (MPA). Besides a basic version of the MPA, three binary variants are developed for this goal, called MPAS, MPAS10 and MPAV. We test the proposed MPA variants with comprehensive comparisons to several optimization algorithms using different evaluation indicators as well as statistical tests to ensure their performance quality. We conclude that MPAV recorded the best performance compared to other MPA variants as well as other compared methods.