Stress detection and alleviation system and method

1. A machine learning based stress detection and alleviation (SoDA) system for a user comprising a SoDA device configured with one or more processors that receive wearable medical sensor (WMS) data from a plurality of WMSs, the processors being programmed to:

preprocess the WMS data from each WMS via range normalization;

extract pluralities of features from the WMS data for each WMS;

remove correlated features from the pluralities of features to obtain a reduced set of features via supervised attribute selection, thresholding, and principal component analysis, wherein:

supervised attribute selection comprises choosing features based on their individual contribution to accuracy and their redundancy with respect to other features;

thresholding comprises removing features that appear less than a predetermined threshold when all data from supervised attribute selection is combined; and

principal component analysis comprises transforming the data from supervised attribute selection into a number of orthogonal variables, the number of orthogonal variables being determined when a classification accuracy stops increasing;

classify the reduced set of features to determine whether the user is stressed by applying a computational operation corresponding to one of a generalized stress model and an individualized stress model, the generalized stress model being based on WMS data obtained from a plurality of users, the individualized stress model being based on WMS data obtained from the user; and

when classification indicates the user is stressed, monitor a set of features from the WMS data that is different from the reduced set of features for a predetermined amount of time in response to a stress mitigation technique.

2. The system of claim 1 , wherein the WMS data comprises electrocardiogram (ECG) data, galvanic skin response (GSR) data, respiration rate data, blood pressure data, and blood oximeter data.

3. The system of claim 1 , wherein classifying the reduced set of features comprises performing binary classification via support vector machine (SVM) or k-nearest neighbor (kNN).

4. The system of claim 1 , wherein the stress mitigation technique comprises one of listening to classical music, practicing micro-meditation, holding a warm stone, and receiving good news.

5. The system of claim 1 , wherein the processors are further programmed to preprocess the WMS data via at least one of:

removing outlier data and replacing the outlier data with at least one of upper and lower thresholds; and

denoising the WMS data with one or more filters.

6. The system of claim 1 , wherein supervised attribute selection further comprises forward feature selection and subset evaluation.

7. The system of claim 1 , wherein for the generalized stress model, the set of features from the WMS data that is different from the reduced set of features comprises R-R interval, heart rate, and ratio of low frequency to high frequency band power (LF/HF) of an electrocardiogram (ECG) signal.

8. The system of claim 1 , wherein for the individualized stress model, the set of features from the WMS data that is different from the reduced set of features comprises R-R interval, heart rate, inhalation duration, and ratio of low frequency to high frequency band power (LF/HF) of an electrocardiogram (ECG) signal.

9. The system of claim 1 , wherein the processors are further programmed to perform majority voting among the set of features from the WMS data that is different from the reduced set of features to indicate whether the stress mitigation technique reduced stress in the user after the predetermined amount of time.

10. The system of claim 9 , wherein the processors are further programmed to monitor the set of features from the WMS data that is different from the reduced set of features for the predetermined amount of time in response to a second stress mitigation technique when the majority voting indicates the user continues to be stressed.

11. The system of claim 9 , wherein the majority voting indicates the stress mitigation technique reduced stress when more than half of a count of the set of features from the WMS data that is different from the reduced set of features show a reduction in stress, wherein for the generalized stress model, showing a reduction in stress comprises at least one of an increase in R-R interval, a decrease in heart rate, and a decrease in a ratio of low frequency to high frequency band power (LF/HF) of an electrocardiogram (ECG) signal.

12. The system of claim 9 , wherein the majority voting indicates the stress mitigation technique reduced stress when more than half of a count of the set of features from the WMS data that is different from the reduced set of features show a reduction in stress, wherein for the individualized stress model, showing a reduction in stress comprises at least one of an increase in R-R interval, a decrease in heart rate, an increase in inhalation duration, and a decrease in a ratio of low frequency to high frequency band power (LF/HF) of an electrocardiogram (ECG) signal.

13. A machine learning based method for stress detection and alleviation (SoDA) for a user of a SoDA device, the SoDA device including one or more processors, the method comprising:

receiving wearable medical sensor (WMS) data from a plurality of WMSs;

preprocessing the WMS data from each WMS via range normalization;

extracting pluralities of features from the WMS data for each WMS;

removing correlated features from the extracted pluralities of features to obtain a reduced set of features via supervised attribute selection, thresholding, and principal component analysis, wherein:

supervised attribute selection comprises choosing features based on their individual contribution to accuracy and their redundancy with respect to other features;

thresholding comprises removing features that appear less than a predetermined threshold when all data from supervised attribute selection is combined; and

principal component analysis comprises transforming the data from supervised attribute selection into a number of orthogonal variables, the number of orthogonal variables being determined when a classification accuracy stops increasing;

classifying the reduced set of features in order to determine whether the user is stressed by applying a computational operation corresponding to one of a generalized stress model and an individualized stress model, the generalized stress model being based on WMS data obtained from a plurality of users, the individualized stress model being based on WMS data obtained from the user; and

when classification indicates the user is stressed, monitoring a set of features from the WMS data that is different from the reduced set of features for a predetermined amount of time in response to a stress mitigation technique.

14. The method of claim 13 , wherein the WMS data comprises electrocardiogram (ECG) data, galvanic skin response (GSR) data, respiration rate data, blood pressure data, and blood oximeter data.

15. The method of claim 13 , wherein classifying the reduced set of features comprises performing binary classification via support vector machine (SVM) or k-nearest neighbor (kNN).

16. The method of claim 13 , wherein the stress mitigation technique comprises one of listening to classical music, practicing micro-meditation, holding a warm stone, and receiving good news.

17. The method of claim 13 , further comprising performing majority voting among the set of features from the WMS data that is different from the reduced set of features to indicate whether the stress mitigation technique reduced stress in the user after the predetermined amount of time.

18. The method of claim 17 , further comprising monitoring the set of features from the WMS data that is different from the reduced set of features for the predetermined amount of time in response to a second stress mitigation technique when the majority voting indicates the user continues to be stressed.

19. A non-transitory computer-readable medium having stored thereon a computer program for execution by a processor configured to perform a machine learning based method for stress detection and alleviation of a user, the method comprising:

receiving wearable medical sensor (WMS) data from a plurality of WMSs;

preprocessing the WMS data from each WMS via range normalization;

extracting pluralities of features from the WMS data for each WMS;

removing correlated features from the extracted pluralities of features to obtain a reduced set of features via supervised attribute selection, thresholding, and principal component analysis, wherein:

supervised attribute selection comprises choosing features based on their individual contribution to accuracy and their redundancy with respect to other features;

thresholding comprises removing features that appear less than a predetermined threshold when all data from supervised attribute selection is combined; and

principal component analysis comprises transforming the data from supervised attribute selection into a number of orthogonal variables, the number of orthogonal variables being determined when a classification accuracy stops increasing;

classifying the reduced set of features in order to determine whether the user is stressed by applying a computational operation corresponding to one of a generalized stress model and an individualized stress model, the generalized stress model being based on WMS data obtained from a plurality of users, the individualized stress model being based on WMS data obtained from the user; and

when classification indicates the user is stressed, monitoring a set of features from the WMS data that is different from the reduced set of features for a predetermined amount of time in response to a stress mitigation technique.

20. The medium of claim 19 , wherein the WMS data comprises electrocardiogram (ECG) data, galvanic skin response (GSR) data, respiration rate data, blood pressure data, and blood oximeter data.

21. The medium of claim 19 , wherein classifying the reduced set of features comprises performing binary classification via support vector machine (SVM) or k-nearest neighbor (kNN).

22. The medium of claim 19 , wherein the stress mitigation technique comprises one of listening to classical music, practicing micro-meditation, holding a warm stone, and receiving good news.

23. The medium of claim 19 , wherein the method further comprises performing majority voting among the set of features from the WMS data that is different from the reduced set of features to indicate whether the stress mitigation technique reduced stress in the user after the predetermined amount of time.

24. The medium of claim 23 , wherein the method further comprises monitoring the set of features from the WMS data that is different from the reduced set of features for the predetermined amount of time in response to a second stress mitigation technique when majority voting indicates the user continues to be stressed.