Data augmentation of time-series data in human movement biomechanics: A scoping review

The integration of machine learning and deep learning methodologies has transformed data analytics in biomechanics. However, the field faces challenges such as limited large-scale data sets, high data acquisition costs, and restricted participant access that hinder the development of robust algorithms. Additional issues include variability in sensor placement, soft tissue artifacts, and low diversity in movement patterns. These challenges make it difficult to train models that perform reliably across individuals, tasks, and settings. Data augmentation can help address these limitations, but its use in biomechanical time-series data remains insufficiently evaluated.

This scoping review on data augmentation for biomechanical time-series data focuses on examining current techniques, evaluating their effectiveness, and offering recommendations for their application.

Four online databases (PubMed, IEEE Xplore, Scopus, and Web of Science) were used to find studies published between 2013 and 2024. Following PRISMA-ScR guidelines, two screening processes were conducted to identify relevant publications.

21 publications were identified as relevant. There is no universal best practice for augmenting biomechanical time-series data; instead, methods vary based on study aims. A key issue identified is the absence of soft tissue artifacts in synthetic data, leading to discrepancies and emphasizing the need for realistic techniques. Furthermore, many studies lack proper evaluation of augmentation methods, making it difficult to understand the effects of different techniques. This understanding is crucial for assessing the impact of the augmented data set on downstream models and evaluating the quality of the data augmentation process.

This review highlights the importance of data augmentation in addressing limited data availability and improving model generalization in biomechanics. Tailoring augmentation to data characteristics can enhance the performance and relevance of predictive models. However, understanding how different augmentation techniques impact data quality and downstream performance remains essential for developing better methods.