Psychological Stress Detection Based on Physiological Signals

心理壓力是現代人許多身心疾病的因素，也是許多社會、經濟問題和個人問題的主因，到醫院就診的病患有超過70%是與心理壓力直接或間接造成的疾病，在美國每年與壓力相關的經濟損失更高達 2000 億美元以上，因此我們希望透過對壓力偵測的研究，可以在未來的日常生活中即時偵測心理壓力狀況，幫助人們及時排解壓力恢復身心健康，從而有好的身心健康狀態以及人生幸福。 在這個研究中我們做了兩個實驗。第一個實驗確立了情境刺激壓力的順序，因此我們得以提供一個較好的刺激方式，可以在實驗室控制壓力刺激的程度。實驗結果建議壓力由小到大：第四象限第一象限第三象限第二象限，這樣的刺激順序較能有效激起壓力反應。同時我們也建立了一套挑選情境內容的方法，用來選擇適當的情境刺激內容當作壓力源。第二個實驗我們用來驗證第一個實驗的結果並找尋較好的壓力分類方式以及適合用來做壓力分類的特徵值。我們使用 SFFS 來進行特徵值選擇，並使用選擇出來較佳的特徵值進行壓力分類，確實經過特徵值選擇後，我們使用較少的特徵值來進行分類就有較佳的正確率。同時我們也發現使用自我評量的方式來給予壓力值，這樣的做法由於人本身的不確定性無法提供太精確的分類，我們把壓力狀態整合成為有壓力/沒壓力兩種分類後，準確率最高可以提高到78%左右，以各種生理訊號的組合平均來說約為68.48%，整體來說提昇了34%。 在本研究中，我們建立了一套完整的系統機制用來進行心理壓力研究，包括如何系統化地挑選進行壓力刺激、進行壓力刺激、收集生理訊號、分析生理訊號以及使用生理訊號辨識和預測心理壓力。由於呼吸是我們唯一可控制的生理訊號，可以用來協助調節壓力和情緒，因此我們建議穿戴式裝置上較好的壓力偵測訊號是皮膚導電度和心跳。Psychological stress is one of the major causes of physical and psychological diseases in the modern world that more than 70% hospital visitors are linked to it. Stress also results more than 200 billion economic loss in U.S. Therefore, we hope the stress research can detect stress in daily life in the future and improve health and well-beings of humankind. We did 2 experiments in this research. Experiment 1 figures out the affective stimuli from relaxed to stressful is Quadrant 4 Quadrant 1 Quadrant 3 Quadrant 2. Experiment 2 proves that the stimuli sequence works. We proposed a method to choose appropriate affective multimedia contents to be stressors of the stress stimuli system. And we also figure out some good features and classification process for stress classification. Feature selection does help in reduce feature numbers of each physiological signals and improve the classification accuracies. Self-assessment is subjective label that can’t provide precise stress labels for classification. After we merge into 2 classes of stress, the maximum accuracy of classification is approximately 78%, improved about 34% in general. In this research, we proposed a completed process to research psychological stress, includes stress stimuli selection, stress stimulation, physiological signal collection, physiological signal analysis, and psychological stress recognition and prediction via physiological signals. Because of respiration is the only physiological signal which we can control by our will to reduce stress and adjust emotion. We recommend GSR and BVP is the best physiological signal to measure stress in wearable devices