INTEGRATIVE APPROACH TO STRESS Detection from DIVERSE ASSESSMENTS

Abstract

Stress detection is now working as a significant working area because it has a deeper impact on the mental and physical health of individuals as well as their productivity. Stress in the past has been measured using different approaches targeting one aspect or facet of data, either biochemical or biomechanical, or cognitive or behavioral. This paper proposes a multimodal approach that involves the use of multiple measurements such as physiological measurements which may include pulse rate, electro dermal activity, facial expressions, speech and self-reported measures. The framework involves these multiple data sources as a unified stream and applies machine learning to improve the reliability and sensitivity of stress classification. Assessing the overall effectiveness of the proposed system, we build our dataset specifically for the experimental setting with tasks stimulating the experience of different levels of stress. Feature extraction and fusion process is designed in such a way that relative strength of each modality is maintained intact. Experimental outcomes show that the proposed multi-modal framework produces better overall accuracy than systems that use only a single mode, in addition to reducing false positive stress identification. This research highlights the possibility of practical applications of the multimodal systems including in a work environment to address stress, interventions for individuals’ mental health, and other forms of human computer interaction. Outside work that is under way, further studies will address its use in dynamic sessions and invalidation of individual scores across various populations and settings.