Research On Mental Stress Recognition Based On Heart Rate Variability

Along with the increasing of human life pressure,psychological stress has become an important factor influencing human's physical and mental health,and may affected the quality of life.Automatically psychological pressure measure is benefited for individual to adjust their own stress status,and this can avoid body and mental stress from harm damage.At the same time,wearable device has been used to monitor cardiac activity in a popularity trend.This helps Heart Rate Variability(HRV)becoming a useable feature to identify psychological pressure automatically.At present,the main difficulties of applying HRV signal to automatically identify the psychological stress are: firstly,no public HRV data set existed for detecting psychological stress,we need to design an experiment to induce stress,synchronically collect the ECG data,and determine the stress tag scientifically.Secondly,removing the noise of ECG signal has always been a difficult in the field of ECG signal acquisition.Thirdly,there is no uniform standard on HRV feature set selection and curtained contribution degree of these features for stress recognition.Finally,we need to choose an appropriate classifier to establish the psychological stress identification model based on HRV features,in order to ensure the accuracy of the identification of psychological stress.In this paper,the mobile phone game-"rhythm master" was chose as the source of stress.According to the game's low,medium and high level,subjects' game parameters(such as hitting,lost,perfect hitting and so on),facial expressions and self-assessments,we set stress label comprehensively.And we recorded the subjects' ECG signal simultaneously.Furthermore,the location of R wave can be identified accurately from ECG signal by wavelet analysis,and at last basic data of HRV was acquired.Information of time domain,frequency domain and nonlinear analysis of HRV were combined to finish HRV feature extraction.In this procedure we extracted features of time domain by statistic method,features of frequency domain by LB periodogram which can overcome HRV signal spectrum's distortion caused by non-uniform sampling.And We also found that there was a certain relationship between the scatter plot's tightness in nonlinear analysis and psychological stress level,from that the relevant nonlinear features were extracted.This paper adopted Random Forests(RF)to build psychological stress identification model based on HRV.According to the characteristic of the RF,we ordered HRV features by psychological stress recognition contribution.Finally,comparisons with K Nearest Neighbor and Logistic Regression stress models,we evaluated the recognition performance of psychological stress model based on RF proposed.Experiments showed that HRV can identify the change of the stress state,especially when the stress was high.At the same time,through the contrast experiments of the three classifier model,RF model has a good recognition performance,especially in high stress level,recognition rate reached beyond 90%.