Research On Description Methods Of Stress State Based On Heart Rate Variability Parameters

In the highly competitive society,the stress is always with us,so the automatic and rapid identification of stress is of great significance.The method of stress emotion recognition based on biological signal is not easy to be camouflaged.And biological signal is widely used to detect the internal and hidden emotions.Heart rate variability(HRV)extracted from Electrocardiogram(ECG)signal,one of biological signal is proved to be used to recognize stress.HRV is consisting of a set of parameters.The parameter values were used as features to support the recognition system.It was generally accomplished by establishing a classification model,such as random forest,logistic regression,and naive bayesian and so on.However,it is impossible for such systems to express the relationship between the different stress states and HRV features can be expressed by quantitative rules.In order to understand the quantitative expression of features in different stress state,decision tree algorithm was introduced in the paper.The feature can be expressed in combination rules,and each feature was approximate quantified.Binary trees were gotten by CART algorithm,and rules were obtained from those trees and the range of the HRV parameters.In order to overcome the redundant rules,the reduced error pruning(REP)algorithm was used to optimize rules.The experiment results showed that the average recognition rate of stress state is 89.4%.After analyzing the wrong rules,it is found that higher error rate of parameter was on boundary,namely rigid boundary values in stress recognition were lack of inclusiveness.In order to increase the ability of parameters inclusive,a strategy was proposed.HRV parameter value was discretized into three parts.Four quantile box plot and normal distribution methods discretized each HRV feature value,by which binary trees were constructed,Experiment results showed that four quantile box plot method was better than the normal distribution,the average recognition rate of stress state was up to 96.1%.