Data preprocessing and machine learning in wearable data analysis: assessing efficacy and challenges for mental health monitoring of healthcare professionals
- Venue
- Université Paris Cité
- Year
- 17 December 2024
- Role
- Author
The question
What can wearable sensor data honestly support when the question is a clinician's mental health?
Abstract
Doctoral thesis examining the efficacy and the limits of data-processing and machine-learning methods applied to wearable device data, for monitoring the mental health of frontline healthcare professionals during COVID-19. The work encodes time-series data using Gramian Angular Field transformation for convolutional models, and reports 97.01% classification accuracy in distinguishing at-risk from non-at-risk cases against clinical thresholds.
Role
Doctoral candidate and sole author. Defended at Université Paris Cité on 17 December 2024, supervised by François Taddei and Harri Ketamo.
Authors. Roy Saurabh
A model that performs well on a cohort can still be the wrong instrument for an individual decision.
Method, and its limits
01
Signals from the body
Wearable devices recorded physiological time series from frontline healthcare professionals during COVID-19.
02
Series become images
The series were encoded with a Gramian Angular Field transformation, so that convolutional models could read temporal structure as texture.
03
A reported result
The thesis reports 97.01% accuracy in distinguishing at-risk from non-at-risk cases against clinical thresholds.
04
What it cannot carry
Sparse, noisy and deeply contextual signals limit what any such figure can justify for one person.
97.01%
Classification accuracy reported in the thesis
A methodological result, attributed to the thesis. It is not a claim about clinical deployment.
Source: theses.fr ↗theses.fr
Thesis record s310267
Prétraitement des données et apprentissage automatique dans l'analyse des données des dispositifs portables : évaluation de l'efficacité et des défis pour le suivi de la santé mentale des professionnels de la santé
Evidence
- 01
theses.fr
Thesis record s310267
Doctorate awarded by Université Paris Cité, defended 17 December 2024, in Mathematics and Computer Science, supervised by François Taddei and Harri Ketamo.
- 02
ABES
Thesis DOI record
Registers the thesis against ORCID 0000-0003-3439-7731 with affiliation Université Paris Cité, confirming that the indexed author name 'Saurabh Roy' is the same researcher.
View original ↗, opens in a new tabDOI ↗, opens in a new tab
Citation
Roy, S. (2024). Data preprocessing and machine learning in wearable data analysis: Assessing efficacy and challenges for mental health monitoring of healthcare professionals [Doctoral dissertation, Université Paris Cité]. https://doi.org/10.70675/bdabcabdz273cz4ff3z8b79zc9434acdf00d
Next project
Data for learning
The Transformative Potential of Data for Learning