Investigating Receptivity and Affect Using Machine Learning: Ecological Momentary Assessment and Wearable Sensing...

Area of Science:

• Mobile health (mHealth)
• Wearable sensor technology
• Machine learning (ML)
• Ecological Momentary Assessment (EMA)

Background:

• Participant receptivity is crucial for mHealth study success, especially when collecting subjective health data.
• Low compliance rates in certain populations can impact data quality.
• ML and sensor data offer innovative ways to optimize survey delivery timing.

Purpose of the Study:

• To investigate the impact of an ML-based EMA delivery system on participants' reported emotional states.
• To identify factors influencing receptivity to EMAs in a wearable sensor-based study.
• To analyze physiological indicators of receptivity and affect, and their interaction.

Main Methods:

• Collected data from 45 healthy participants using wearable sensors (electrodermal activity, accelerometer, ECG, skin temperature).
• Administered 10 EMAs daily assessing perceived mood.
• Employed unsupervised (k-means clustering) and supervised (random forest, neural networks) ML to infer affect and receptivity during non-responses.

Main Results:

• Triggering EMAs via a receptivity model decreased self-reported negative affect by over 3 points (0.29 SDs).
• Predicted affect during non-responses showed a bimodal distribution, indicating EMAs were more frequently initiated during positive emotional states.
• A significant relationship was observed between affect and receptivity.

Conclusions:

• The relationship between affect and receptivity can influence mHealth study efficacy, particularly with ML-triggered EMAs.
• Future research should aim for smart triggers that enhance EMA receptivity without biasing emotional state reporting.