1. Smartphone-based Activity Recognition for Pervasive Healthcare - Utilizing Cloud Infrastructure for Data Modeling Bingchuan Yuan, John Herbert University College Cork, Ireland

2. Outline 2 Introduction 1 Activity Recognition Approach 2 Cloud-based Data Modeling 34 Conclusion 5 Experiment & Result

3. Introduction  Pervasive Healthcare Traditional clinical setting  Home-centered setting Wireless Sensor Networks (WSNs) &Communication technologies 3 WSN Internet

4. Introduction  CARA for Pervasive Healthcare CARA (Context-Aware Real-time Assistant) Real-time Intelligent At-home healthcare  Activity Recognition in CARA Activity of Daily Living (ADL) monitoring Anomaly detection 4

5. Introduction  State of The Art - Environmental sensor-based approach Pros: ambient assistant monitoring Cons: intrusive, large installation - Wearable sensor-based approach Pros: small, low cost, non-invasive Cons: customized, impractical, processing power - Smartphone-based approach Pros: ubiquity, sensing and computing Cons: battery, insufficient accuracy 5

6. Activity Recognition  Our Approach Smartphone-based Wearable wireless sensor integrated Hybrid Classifier Cloud-based data modeling 6

7. Activity Recognition  ADLs in a Home Environment - Static Posture: Sitting, Standing, Lying, Bending and Leaning back - Dynamic Movement: Walking, Running, Walking Stairs, Washing Hands, Sweeping and Falling 7

8. Activity Recognition  Overview 8 Load Classification Model Data Collection Feature Extraction Distinguish Static and Dynamic Activity Activity Classification Classification Model Optimization

9. Activity Recognition  Feature Extraction 9 WalkingRunningSweeping Washing Hand 1s - Window

10. Activity Recognition  Feature Extraction 10 FeatureTrunk Acceleration Thigh Acceleration Thigh Orientation MinX, Y, Z, |ACC| Azimuth, Pitch, Roll, |GYRO| MaxX, Y, Z, |ACC| Azimuth, Pitch, Roll, |GYRO| MeanX, Y, Z, |ACC| Azimuth, Pitch, Roll, |GYRO| Standard Deviation X, Y, Z, |ACC| Azimuth, Pitch, Roll, |GYRO| Zero CrossX, Y, Z Azimuth, Pitch, Roll Mean Cross|ACC| |GYRO| AngularX, Y, Z

11. Activity Recognition  Distinguish static and dynamic Activity 11 Dynamic Activity Static Activity

12. Activity Recognition  Real-time Activity Classification Using Hybrid Classifier - Static activity: Threshold-based method - Dynamic activity: Machine learning classification model 12

13. Activity Recognition 13 Inclination Angle:  Static Activity

14. Activity Recognition  Dynamic Activity Weka* for data mining Machine learning algorithms: - Bayesian Network - Decision Tree - K-Nearest Neighbor - Neural Network 14 * Weka 3: Data Mining Software (Developed by University of Waikato)

15. Activity Recognition  Transition of Activity States S 0 : Transitional State S 1 -S 5 : State of each activity R: Transition Rule 15

16. Cloud-based Data Modeling  Activity Data Modeling Training the classification models: tradeoff between accuracy and cost -Personalized model: One for each individual (better accuracy) -Universal model: One size fits all (lower cost) 16

17. Cloud-based Data Modeling  Model Adaptation 17

18. Cloud-based Data Modeling  Cloud-based Data Analysis Framework 18

19. Experiment and Result  Data Collection Eight volunteers Home setting Activity tasks Supervised learning Ground truth testing set 19

20. Experiment and Result  Data Set Activity instances of the Default Model 20

21. Experiment and Result  Confusion Matrix Table (Default Model) 21 ActivityabcdefghijklACC WALKING (a)185201000060000099.14% RUNNING (b)111053210000000097.01% WALK STAIRS (c)470193740014002097.09% SWEEPING (d)601137814040000098.22% WASHING HANDS (e)0000873070021098.87% FALLING (f)002123261530061.54% STANDING (g)00000082200000100% SITTING (h)00000097200000100% LYING (i)0000000164900099.85% BENDING (j)00000000071800100% LEANING BACK (k)00000000005570100% ROLLING (l)00410111580118785.78% *Default Model built by the KNN classifier and evaluated using 10-fold cross-validation

22. Experiment and Result  Performance Overview 22 Overall model accuracy for the female user B Overall model accuracy for the male user A

23. Experiment and Result 23 ClassifierTP RateFP RatePrecisionRecallF-ScoreTime(ms)Accuracy First Run (1980 instances  1874 instances) Decision Tree0.6840.0340.0890.6840.657183868.37% Bayesian Network0.8450.0110.9310.8450.860272584.45% K-Nearest Neighbor0.7240.0440.8110.7240.684484972.35% Neural Network0.7420.0400.8110.7420.7208468274.23% Second Run (3493 instances  3392 instances) Decision Tree0.9580.0060.9600.9580.957124895.80% Bayesian Network0.8970.0100.9430.8970.899148289.67% K-Nearest Neighbor0.7950.0360.8600.7950.763550479.51% Neural Network0.9700.0040.9720.970 14511197.04% Third Run (5482 instances  5403 instances) Decision Tree0.9660.0060.9670.966 135896.61% Bayesian Network0.9550.0050.9670.9550.958172795.48% K-Nearest Neighbor0.8280.0310.8730.8280.810701982.84% Neural Network0.9850.0020.985 22777498.49%

24. Conclusion  Key Points Smartphone-based Wearable wireless sensor integrated Hybrid Classifier Cloud-based data modeling  Future Work Automatically distinguish static and dynamic activity Dynamically allocate system resource in the cloud 24

25. University College Cork