Decoding Emotional Valence from Wearables: Can Our Data Reveal Our True Feelings?

TL;DR

The paper investigates whether consumer-grade wearables can detect emotional valence in real-world settings by linking PPG-derived HRV features to self-reported valence. Using a two-week study with 15 participants and PANAS-10 assessments, the authors extract HRV metrics around self-reported states and perform a valence classification, achieving an F1 of 0.65 for high vs low positive affect. They report meaningful correlations between HRV features (e.g., IBI, BPM, SDNN, SD2) and positive affect, while negative affect shows fewer robust associations. The study demonstrates feasibility of wearable-based emotion detection and discusses ethical considerations, data limitations, and directions for larger-scale validation and model improvement for mobile mental health interventions.

Abstract

Automatic detection and tracking of emotional states has the potential for helping individuals with various mental health conditions. While previous studies have captured physiological signals using wearable devices in laboratory settings, providing valuable insights into the relationship between physiological responses and mental states, the transfer of these findings to real-life scenarios is still in its nascent stages. Our research aims to bridge the gap between laboratory-based studies and real-life settings by leveraging consumer-grade wearables and self-report measures. We conducted a preliminary study involving 15 healthy participants to assess the efficacy of wearables in capturing user valence in real-world settings. In this paper, we present the initial analysis of the collected data, focusing primarily on the results of valence classification. Our findings demonstrate promising results in distinguishing between high and low positive valence, achieving an F1 score of 0.65. This research opens up avenues for future research in the field of mobile mental health interventions.

Figures (3)

• Figure 1: An overview of our data acquisition flow. The participants could fill in the PANAS questionnaire either on the Android smartphone or Samsung Galaxy Watch 4 (preferred option). The completed surveys were sent to the DynamoDB database. During the survey completion on the smartwatch, the PPG signal was collected and stored in the local database. The signal was then transformed and saved in .csv files which were later uploaded to the S3 bucket.
• Figure 2: Distribution of self-reported responses.
• Figure 3: The figure shows the Pearson correlations computed for the self-reported emotional scores and the features extracted from the blood volume signal. The correlations that were found to be statistically significant (p < 0.05) are shown in colour, with red corresponding to the positive correlations and blue corresponding to the negative correlations. The upper left corner shows how the relationship between the self-reported emotions, including the cognitive load and the calculated overall positive and negative affect. The bottom right corner focuses on the relationships between the features extracted from the blood volume signal. The area within the dotted lines shows the correlations between the emotional scores and the signal obtained from the wearable device.