2019/10/03 by Kizito Nkurikiyeyezu, Nkurikiyeyezu, Kizito, Anna Yokokubo +3 · 1 citation
Engineering · Psychology · #Emotion and Mood Recognition #FOS: Computer and information sciences #Human-Computer Interaction (cs.HC) #Mental Health Research Topics #Non-Invasive Vital Sign Monitoring
paper · pdf · doi:10.48550/arxiv.1910.01770
openalex publication_date 2019/10/03 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
Because stress is subjective and is expressed differently from one person to\nanother, generic stress prediction models (i.e., models that predict the stress\nof any person) perform crudely. Only person-specific ones (i.e., models that\npredict the stress of a preordained person) yield reliable predictions, but\nthey are not adaptable and costly to deploy in real-world environments. For\nillustration, in an office environment, a stress monitoring system that uses\nperson-specific models would require collecting new data and training a new\nmodel for every employee. Moreover, once deployed, the models would deteriorate\nand need expensive periodic upgrades because stress is dynamic and depends on\nunforeseeable factors. We propose a simple, yet practical and cost effective\ncalibration technique that derives an accurate and personalized stress\nprediction model from physiological samples collected from a large population.\nWe validate our approach on two stress datasets. The results show that our\ntechnique performs much better than a generic model. For instance, a generic\nmodel achieved only a 42.5% accuracy. However, with only 100 calibration\nsamples, we raised its accuracy to 95.2% We also propose a blueprint for a\nstress monitoring system based on our strategy, and we debate its merits and\nlimitation. Finally, we made public our source code and the relevant datasets\nto allow other researchers to replicate our findings.\n