A Deep Ensemble model to Recognize Human Activities using inertial sensors on smartphones

Human Activity Recognition (HAR) is a crucial area of research in the fields of health and human-machine interaction. Smartphones have emerged as a popular choice for HAR due to their ubiquitous nature in daily life. Most available HAR datasets are collected in laboratory settings, which do not accurately represent real-world scenarios. To address this limitation, we collect a real-world dataset using smartphone inertial sensors from 62 individuals. Our collected dataset is small, noisy, and has variable frequency, which add complexity to the activity recognition process. In this paper, we propose a novel ensemble of hybrid deep models for HAR using smartphone sensors, which improves generalization performance and outperforms current methods with an accuracy of 93.9. The implications of these findings are significant for the development of reliable and accurate HAR systems that can operate effectively in real-world scenarios characterized by high intra-class diversity and inter-class similarity.