摘要
针对目前心理压力问题比较严重,收集生理数据评估心理压力存在成本高、主观性强等问题,提出了一种新的基于手机数据的压力评估方法BSTL+XGDT(Borderline1 SMOTE Tomeklinks+eXtreme Gradient Boosting),将压力水平精确划分为5个级别.首先从手机数据提取特征生成样本,对样本进行BSTL采样,然后用XGDT过滤特征和RFE(Recursive Feature Elimination)筛选特征,同时,利用采样前后的数据及特征筛选前后的数据训练XGDT、支持向量机(SVC)、随机森林(RF)、K近邻(KNN)、决策树(DT)、多层感知机(MLP)、标签传播(LS)方法,结果显示方法BSTL+XGDT优于其他方法.
In view of the serious problem of psychological stress at present,and the problems of high cost and strong subjectivity in collecting physiological data to assess psychological stress,this paper proposed a new stress assessment method based on mobile phone data,BSTL+XGDT(Borderline1 SMOTE Tomeklinks+eXtreme Gradient Boosting),which could divide the stress levels into 5 levels accurately.First,the features were extracted from mobile phone data to generate samples,and BSTL was used to balance samples.Then,XGDT and RFE methods were combined to select feature,meanwhile,the data before and after sampling and the data before and after feature selection were used to train XGDT,Support Vector Machine(SVC),Random Forest(RF),K Nearest neighbor(KNN),Decision tree(DT),Multi-layer perceptual(MLP),label propagation(LS).The experimental results show that the method BSTL+XGDT outperforms the other methods.
作者
孙永明
杨进
SUN Yongming;YANG Jin(School of Science, University of Shanghai for Science and Technology, Shanghai 200093, China)
出处
《经济数学》
2020年第4期148-158,共11页
Journal of Quantitative Economics
基金
国家教育部人文社科规划基金项目(16YJA630037)
上海市一流学科建设项目(S1201YLXK)。
