Using Thermal and Contact Sensors for Mood Detection in Smart Living Environments

Mental health in everyday life can be supported using ambient and wearable sensors, which could detect when a person is feeling well (e.g. happy or calm) or not feeling well (e.g. stressed or sad). Identifying positive or negative moods can be useful for interventions to reinforce or improve them respectively. Previous work that used data from wearable sensors (accelerometers) in a personalized way successfully identified mood while users performed Activities of Daily Living (ADLs). This paper presents a general approach to using data from ambient (contact and thermal) sensors to identify the mood of users while performing ADLs. The rationale for using ambient sensors is their non-intrusiveness. Users do not have to be as concerned about their use or maintenance as with wearable sensors. Data was collected from 15 participants performing ADLs in 7 sessions. Accuracy classification results obtained with 7 algorithms underperformed, with the highest value being 45.74% with Random Forest in 10-Fold Cross Validation. For future work, data from other sensors (in addition to the accelerometer) will be collected to support the improvement of the personalized approach presented, and other evaluation metrics (e.g. F1-score and AUC) will be used.