Accelerometer Data Harmonization and Feature Engineering for Gait Analytics
- 1Heilbronn University: Hochschule Heilbronn
Abstract
Gait analytics is utilized for a wide range of health-related applications, ranging from fitness tracking to medical diagnostics, and beyond. Inertial measurement units that are ubiquitously available in mobile devices provide an accessible means of gait analysis but differences in sensor specifications, placement, and recording protocols limit comparability across studies. This study developed and evaluated an analysis-centered data warehouse and standardized quality-assessed processing pipeline to facilitate the integration and joint analysis of heterogeneous gait accelerometer data. Six publicly available datasets including a total of 935 persons were integrated into a star-schema data warehouse, and feature comparability was assessed across datasets and sensor locations. Gender classification served as a proof-of-concept downstream analysis. Gait-cycle extraction, which serves as foundation for feature extraction, was successfully achieved with an accuracy above 99% for the dataset that provides ground truth information. Comparability varied across datasets and sensor configurations, with correlation coefficients between average gait cycles ranging from 0.5 to close to 1.0. Centrally positioned sensors generally show greater correspondence than peripheral sensors, although differences in statistical features persist even between sensors at nominally identical locations. Gender classification performance remains largely near chance level, and no consistent benefit from pooling datasets could be established. These findings demonstrate that heterogeneous gait accelerometer data can be integrated and processed within a common analytical framework, but standardized processing alone does not ensure comparability of the resulting features. The proposed open-source framework implemented in Python provides a foundation for reproducible cross-study gait analysis and for investigating residual heterogeneity associated with datasets and sensor configurations.
Related
- Benchmarking of bulk transcriptomic harmonization tools in a multi-platform B-cell lymphoma cohort identifies feature-specific quantile normalization and surrogate variable analysis as top-performing methods
- Feature harmonization reveals that apparent cross-cohort differences in tau-positivity prediction reflect variable availability and cohort stage structure
- Optimized Feature Engineering and Machine Learning Approaches for Early Prediction of Renal Cell Carcinoma
- Development of a parallel cascaded adaptive ensemble framework with feature engineering process for the prediction of neurodegenerative disease using temporal clinical data
- A novel negative-based CNN and deep feature engineering framework for ACL rupture detection from knee MRI
BibTeX
@misc{carrle2026accelerometer,
title = {{Accelerometer Data Harmonization and Feature Engineering for Gait Analytics}},
author = {F. P. Carrle and F. Schmidt and A. Reichenbach},
howpublished = {medRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.05.26367336},
url = {https://www.medrxiv.org/content/10.64898/2026.11.05.26367336v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Accelerometer Data Harmonization and Feature Engineering for Gait Analytics
AU - F. P. Carrle
AU - F. Schmidt
AU - A. Reichenbach
PY - 2026
DA - 2026/11/05
PB - medRxiv
DO - 10.64898/2026.11.05.26367336
UR - https://www.medrxiv.org/content/10.64898/2026.11.05.26367336v1
AB - Gait analytics is utilized for a wide range of health-related applications, ranging from fitness tracking to medical diagnostics, and beyond. Inertial measurement units that are ubiquitously available in mobile devices provide an accessible means of gait analysis but differences in sensor specifications, placement, and recording protocols limit comparability across studies. This study developed and evaluated an analysis-centered data warehouse and standardized quality-assessed processing pipeline to facilitate the integration and joint analysis of heterogeneous gait accelerometer data. Six publicly available datasets including a total of 935 persons were integrated into a star-schema data warehouse, and feature comparability was assessed across datasets and sensor locations. Gender classification served as a proof-of-concept downstream analysis. Gait-cycle extraction, which serves as foundation for feature extraction, was successfully achieved with an accuracy above 99% for the dataset that provides ground truth information. Comparability varied across datasets and sensor configurations, with correlation coefficients between average gait cycles ranging from 0.5 to close to 1.0. Centrally positioned sensors generally show greater correspondence than peripheral sensors, although differences in statistical features persist even between sensors at nominally identical locations. Gender classification performance remains largely near chance level, and no consistent benefit from pooling datasets could be established. These findings demonstrate that heterogeneous gait accelerometer data can be integrated and processed within a common analytical framework, but standardized processing alone does not ensure comparability of the resulting features. The proposed open-source framework implemented in Python provides a foundation for reproducible cross-study gait analysis and for investigating residual heterogeneity associated with datasets and sensor configurations.
ER -