Short answer

When developing AI-driven features for wearables, especially those related to health monitoring, incorporate data collected from a wide range of real-world user scenarios, not just controlled laboratory conditions.

Field
Human Factors
Source
Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (2025)
Method
Comparative evaluation of foundation models
Sample
120 participants
Evidence
Strong effect

Foundation models for photoplethysmography (PPG) trained on diverse, uncurated field data demonstrate superior generalization and performance in real-world wearable applications compared to those trained solely on clinical data. This human factors research insight is drawn from a 2025 study published in Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies. Using Comparative evaluation of foundation models with 120 participants, researchers explored how this design variable affects real-world outcomes. The key design takeaway: When developing AI-driven features for wearables, especially those related to health monitoring, incorporate data collected from a wide range of real-world user scenarios, not just controlled laboratory conditions.

Study
Human FactorsNew This WeekStrong effect

Field-Trained PPG Models Enhance Wearable Health Monitoring Robustness

Foundation models for photoplethysmography (PPG) trained on diverse, uncurated field data demonstrate superior generalization and performance in real-world wearable applications compared to those trained solely on clinical data.

Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · 2025

01

Key Findings

  • 01Pulse-PPG, trained on uncurated field data, shows superior generalization and performance across clinical and mobile health applications in both lab and field settings.
  • 02Pre-training on field data outperforms pre-training on clinical data for many tasks, underscoring the value of real-world data variability.
  • 03Models trained on field data learn more robust representations due to exposure to real-world variability.
02

Application

Design takeaway

When developing AI-driven features for wearables, especially those related to health monitoring, incorporate data collected from a wide range of real-world user scenarios, not just controlled laboratory conditions.

How to apply

When designing a new wearable device that relies on PPG for health tracking, ensure the underlying AI models are trained on datasets that reflect the typical usage patterns and environmental conditions of the target users.

Project actions

  • 01When collecting data for your design project, try to simulate or include variations that mimic real-world use, even if it's challenging.
  • 02Consider how environmental factors (e.g., movement, light, temperature) might affect sensor readings and how your design can account for this.
03

Method & Evidence

AimCan a PPG foundation model trained on uncurated field data generalize better to diverse lab and field settings than models trained on curated clinical data?
MethodComparative evaluation of foundation models
ProcedureA new PPG foundation model (Pulse-PPG) was trained exclusively on raw PPG data collected over 100 days from 120 participants in a field study. This model's performance was then compared against existing state-of-the-art PPG foundation models trained on clinical data across various clinical and mobile health applications in both lab and field settings.
Sample120 participants
ContextWearable biosignal monitoring, mobile health applications

Variables

IVTraining data source (field vs. clinical)
DVGeneralization and performance across lab and field settings
CVType of PPG signal, specific health applications evaluated, evaluation metrics
04

Strengths & Limitations

Strengths

  • +Use of a large, diverse field dataset over an extended period.
  • +Open-sourcing of the model encourages further research and validation.

Limitations

It can be difficult and expensive to collect large-scale, diverse, real-world data for a design project. Ethical considerations and data privacy are also significant challenges.

Reliability & validity

The study's validity is supported by extensive evaluations across multiple settings and comparisons with state-of-the-art models. Reliability is enhanced by the large sample size and long data collection period. The open-source nature allows for independent verification.

Think critically

How might the 'uncutted' nature of field data, while beneficial for robustness, introduce challenges in identifying specific physiological events or anomalies that might be more easily detected in cleaner clinical datasets?

05

Design Principles

"Embrace data diversity and real-world variability in model training for robust human-centric applications."

This research highlights the critical need to incorporate real-world variability into the training of biosignal processing models. For designers of wearable health devices, this means that models trained in controlled lab environments may not perform optimally when deployed in the unpredictable conditions users experience daily, impacting the reliability and utility of health insights.

06

What This Means for Your Design

If you're building a health tracker for a watch, make sure the software learns from data collected by people in their normal lives, not just from people hooked up to machines in a lab. This makes the tracker more accurate when people actually use it.

How to use in your project

  • 1.Reference this study when discussing the importance of data collection methods and their impact on the performance of user-facing technology in your design project.
07

Add to My Project

08

Quick Cite

Paragraph starter

The development of robust biosignal processing models for wearable applications is significantly influenced by the nature of the training data. Research by Saha et al. (2025) demonstrates that foundation models for photoplethysmography (PPG) trained on uncurated field data exhibit superior generalization and performance in real-world settings compared to those trained on curated clinical data. This highlights the critical need for designers to incorporate real-world variability into their data collection and model training strategies to ensure the reliability and effectiveness of user-facing technologies.

09

Source

Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies

Pulse-PPG: An Open-Source Field-Trained PPG Foundation Model for Wearable Applications across Lab and Field Settings

journal · 2025

View source

Questions About This Research

What does the research say about field-trained ppg models enhance wearable health monitoring robustness?
When developing AI-driven features for wearables, especially those related to health monitoring, incorporate data collected from a wide range of real-world user scenarios, not just controlled laboratory conditions. Evidence: Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies (2025).
Why does "Field-Trained PPG Models Enhance Wearable Health Monitoring Robustness" matter for design?
This research highlights the critical need to incorporate real-world variability into the training of biosignal processing models. For designers of wearable health devices, this means that models trained in controlled lab environments may not perform optimally when deployed in the unpredictable conditions users experience daily, impacting the reliability and utility of health insights.
How can designers apply this research?
When developing AI-driven features for wearables, especially those related to health monitoring, incorporate data collected from a wide range of real-world user scenarios, not just controlled laboratory conditions.
What were the main findings?
Pulse-PPG, trained on uncurated field data, shows superior generalization and performance across clinical and mobile health applications in both lab and field settings.. Pre-training on field data outperforms pre-training on clinical data for many tasks, underscoring the value of real-world data variability.. Models trained on field data learn more robust representations due to exposure to real-world variability.
What research method was used?
Comparative evaluation of foundation models with 120 participants.
How strong is the evidence?
Evidence strength is rated Strong effect, based on a 2025 journal from Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies.
What should I do differently in my next project?
When designing a new wearable device that relies on PPG for health tracking, ensure the underlying AI models are trained on datasets that reflect the typical usage patterns and environmental conditions of the target users.
What are the limitations?
The study focuses on PPG signals; findings may not directly translate to other biosignals. The specific characteristics of the 'field' environment and participant demographics could influence generalizability.