Temporal Data Augmentation techniques for sensor-based Human Activity Recognition
Youssef Errafik, Younes Dhassi, Adil Kenzi · 2024
With the advances in generative models, it is becoming crucial to exploit their performance in the analysis of temporal data. Unlike discriminative models, which focus on distinguishing between different classes, generative models focus on representing the distribution of data and their ability to generate new data. However, these models present added complexity due to the need to model the joint distribution in order to generate realistic data. This complexity is even higher for temporal data. Therefore, the scientific community has experimented with different approaches to temporal data augmentation. This in-depth analysis explores the current state of knowledge in this field, providing a summary of the algorithms available and presenting a comparative assessment of existing approaches. We assess the performance of these temporal data augmentation techniques based on their overall impact on the Time-Series Classification (TSC), using metrics specifically in the context of human activity recognition (HAR) on datasets dedicated to this domain.