基于XGBoost的多种生理信号评估心理压力等级方法

基于生理信号客观评估心理压力状态成为目前的研究热点，但最佳评估算法有待进一步探索. 本文选择心算任务诱发受试者的心理压力，采集了21位在校大学生的脑电、心电、皮肤电导、脉搏波4种生理信号. 提取各生理信号时域和频域的多种特征，使用方差分析(ANOVA)、最大相关最小冗余(mRMR)、单个特征支持向量机(SVM)分类准确率、随机森林(RF)特征重要性、梯度上升决策树(GBDT)特征重要性、极端梯度提升(XGBoost)特征重要性6种特征选择方法筛选出有效特征，利用SVM、K近邻(KNN)、高斯朴素贝叶斯(GNB)、自适应提升算法(Adaboost)、GBDT、XGBoost 6种分类器对提取的特征进行分类. 结果得出，GBDT特征筛选与XGBoost分类器的组合模型对心理压力的等级评估效果最佳. 

Psychological Stress Assessment Using Multiple Physiological Signals Based on XGBoost

Objective assessment of psychological stress using physiological signals has become a current research hotspot, but the best algorithm needs to be further explored. In this study, a mental arithmetic task was conducted to induce psychological stress in subjects. Four physiological signals including EEG, ECG, skin conductance, and pulse wave were collected from 21 university students. The features of the time and frequency domains for physiological signals were extracted. Six methods including ANOVA, mRMR, Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost) were utilized to select effective features. SVM, K-Nearest Neighbor (KNN), Gaussian Naive Bayesian (GNB), Adaptive Boosting (Adaboost), GBDT, and XGBoost were conducted to classify the extracted features. The results show that the combined model of GBDT feature selection and XGBoost classifier is the most effective for the assessment of psychological stress on different levels.