Use of Thermal Sensor Data for Personalised Mood Detection in Activities of Daily Living (ADLS)

Abstract

Ambient sensors have been typically used in Human Activity Recognition (HAR) to monitor the activities of people and to detect unusual activities that may affect a person’s wellbeing. The main advantages of ambient sensors are that they are not intrusive and do not require the user to charge them periodically. Thermal sensors are a type of ambient sensor that provides temperature data from the environment in which they are placed, allowing to identify a thermal representation of elements that produce heat, such as people, animals or hot objects. In most cases, the focus of HAR research is on the physical health of people, not on their mental health. This paper presents an investigation on the use of thermal sensor data from people performing Activities of Daily Living (ADLs) to identify mood in a personalised way. Thermal data was collected from 15 participants performing the ADLs of preparing and drinking a hot beverage in 7 sessions. At the start of each session participants reported their mood. Classification results were produced for each participant using the Support Vector Machines (SVM) model in 10-Fold Cross Validation (CV) and in 80/20 split. The average accuracy values obtained of 0.9123 (80/20) and 0.9233 (CV), and of Cohen’s Kappa Coefficient of 0.8375 (80/20) and 0.8574 (CV) are promising for a thermal sensor personalised mood detection approach.