ElderHAR - Deep-Learning Framework for Elder Inertial Data Activity Recognition
Year: 2026
Type: software
Abstract: Overview This record contains a versioned reproducibility package for ElderHAR, a deep learning framework for inertial sensor-based activity recognition in older adults. The package supports reproducible data preprocessing, subject-independent model development, held-out evaluation, statistical analysis, sensor ablation, and post-hoc explainability. The archived version includes the source code, frozen participant splits, experiment configurations, model definitions, analysis scripts, automated tests, and documentation required to reproduce the principal computational analyses. Raw participant data and trained model checkpoints are not included. The ElderHAR dataset is available separately through Zenodo (DOI: https://doi.org/10.5281/zenodo.17191997). Citation If you use this software, its components, or the accompanying experimental protocol, please cite: this Zenodo software record using its version-specific DOI; and the associated peer-reviewed article. The complete article citation and DOI will be added to this record after the article is accepted for publication. Until then, users should cite the software record directly using the citation information provided by Zenodo. Reproducibility scope This archive represents the software version associated with the reported experiments. Subsequent development versions may introduce additional functionality or methodological changes and should not be assumed to reproduce the same results.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7214943142