Activity recognition in health field

BY
YOGESH K M
Research Scholar
EXPLORATION OF ADVANCED DATA MINING MODELS FOR
KNOWLEDGE DISCOVERY FROM mHealth DATASET
(DATA SCIENCE APPROACH)

1. Problem Statement
2. Introduction
3. Motivation of mHealth Applications
4. Literature Review on Related Work
5. Objectives of the Proposed Research
6. Research Methodology
7. Research Plane with Time Schedule
8. Work Under Progress
9. Conclusion
10. References

vital
signs
various human physical activities
data science approach.

Mobile smart devices are becoming increasingly sophisticated. The latest generation of smart
cell phones now incorporates many diverse and powerful smart sensors
Friday, August 12, 2016
 GPS (Geographical position system ) sensor
 Magnetometer sensor
 Gyro meter sensor
 light sensor
 Temperature sensor
 Direction sensor
 Orientation sensor
the latest wearable sensors called mobile health (mHealth)
sensors for health related activities measurements and basic
vital sign of human. Deployment of these diverse mHealth
sensors at different spatial locations on human body sets up
human activity environment where different human activities
recorded and measured can be used for Human Activity
Recognition (HAR) for all basic function of human
(VITAL SIGN).

WHAT IS mHealth..?
 mHealth is the use of mobile and wireless
technologies, such as mobile phones, patient
monitoring devices, personal digital assistants,
and mobile software applications (apps),to
support the achievement of health objectives
 The adoption of mHealth seeks to take
advantage of the explosion in mobile devices
available worldwide

mHealth Applications and Benefits
The United Nations has identified seven application categories within the mHealth field, including:
 Remote monitoring
 Remote data collection
 Education and awareness
 Diagnostic and treatment support
 Disease and epidemic outbreak tracking
 Helpline
Benefits include the ability to:
 Access healthcare information.
 Diagnose & track disease.
 Gather actionable public information.
 Deliver medical education & training

What is physical activity?
Physical activity simply means movement of the body that uses energy.
 Climbing Stairs,
 Cycling,
 Front elevation of arm
 Jogging, jump front andBack
 Knee bending
 Lying down
 Running, Sitting & relaxing
 Standing still
 Waist bend forward,
 Walking.

Activity recognition in health field

Physical activity is important for everyone, but how much you need
depends on your age.
ADULTS (18-60 YEARS):

2. CHILDREN AND ADOLESCENTS (6-17 YEARS)
 Most of the 60 minutes should be either
moderate- or Vigorous intensity aerobic physical activity,
 Should include vigorous-intensity physical activity at least
3 days a week.
 Children and adolescents should do 60 minutes
3. YOUNG CHILDREN (2-5 YEARS)
 There is not a specific recommendation for the number of
minutes young children should be active each day.
 Children ages 2-5 years should play actively several times
each day.
 Their activity may happen in short bursts of time and not
be all at once.
 Physical activities for young children should be
developmentally appropriate, fun, and offer variety

4. LITERATURE REVIEW on Related Work

Article Key concept Limitations
Kwapiz et al [2011] Sensor data mining, activity
recognition Cell phone,
accelerometer
Activity Recognition using cell phone
Accelerometer.
Bingchuan et al
[2014]
ADLs, Smartphone, Wearable
Wireless Sensor, Machine Learning,
Cloud Infrastructure, Unsupervised
Learning, Real-time Activity
Recognition
Smartphone-based Activity Recognition Using
Hybrid Classifier
Utilizing Cloud Infrastructure For Data Analysis.
Torres-Huitzil [2015] Human activity recognition,
accelerometer, smart phone,
mHealth, time domain features
Wireless device used support mHealth services
Vincent S Tseng et
al [2008]
Data mining, Electrocardiogram
analysis, Patient monitoring system,
Vital sign analysis
Vital Sign Data Mining System for Chronic Patient
Monitoring.

Human Activity Recognition based on
Mobile Sensor Data Sets

Relevant classification methods implemented for Human Activity recognition for Mobile health
Monitoring System
Kwapisz, Jennifer R., Gary M. Weiss, and Samuel A. Moore.(2011) "Activity
recognition using cell phone accelerometers." ACM SigKDD Explorations
Newsletter 12, .2, pp.74-82.
Activity
Algorithms Accuracy(%)
J48 M P L P STRAW MAN
Walking 89.9 91.6 93.6 37.2
Jogging 96.5 98.0 98.0 29.2
Upstairs 59.3 61.5 27.5 12.2
Down stairs 55.5 44.3 12.3 10.0
sitting 95.7 92.95.0 92.2 6.4
Standing 93.3 91.9 87.0 5.0
Overall 85.1 91.7 78.1 37.2

Monitoring System
Bingchuan., Herbert, J., & Emamian, Y. (2014). Smartphone-based activity recognition using hybrid classifier. In Proceedings
of the 4th International Conference on Pervasive and Embedded Computing and Communication Systems (PECCS 2014)
Activity
Overall Model accuracy for Female user
Bayes Network Decision Tree K-NN Neural Network
FEMALE
USER
1st 84.45 68.37 72.35 95.97
2nd 89.67 79.51 79.51 96.54
3rd 95.48 96.61 82.84 88.43
Overall Model accuracy for the Male and Female user
1st 84.38 90.74 91.23 87.83
MALE
2nd 92.57 93.02 94.14 90.69
3rd 92.93 94.77 94.41 91.86

Monitoring System
3 Torres-Huitzil et al,(2015) "Accelerometer-Based Human Activity Recognition in Smartphone’s for
Healthcare Services." Mobile Health. Springer International Publishing, pp.147-169.
Activity NB K-NN SVM
Precision Recall Precision Recall Precision Recall
Static 100 99.36 100 99.36 100 99.36
Running 100 100 100 100 100 100
Walking 91.03 85.71 79.27 84.42 91.00 59.09
UP-stairs 80.98 88.69 78.07 86.90 57.30 93.45
Down stairs 82.28 79.27 79.41 65.85 57.52 39.63
Average 90.85 90.60 87.35 87.30 81.16 78.30

HUMAN ACTIVITY RECOGNITION AND ECG VITAL SIGNS

VincentS Tseng et al(2008), “Development of a Vital Sign Data Mining Systemfor Chronic PatientMonitoring”,Proceedingsof
International Conference on Complex,Intelligentand SoftwareIntensive Systems,2008. CISIS 2008,4-7 March2008,pp. 649 - 654.
Sl.No Methods Recall
(True Positive)
Precision
(False Positive)
1 Naïve Bayes Simple 68% 67%
2 Naïve Bays 57% 56%
3 libSVM 61% 56%
4 Radial Bias Function 57% 67%
5 Random-Forest Tree 39% 21%
6 J48 32% 26%
7 1-NN classifier 29% 26%
8 Logistic Regression Tree 25% 36%
9 Classification tree 21% 19%
10 Probit-Regression Tree 21% 35%
11 ADTree 18% 25%
12 Random Forest 7% 4%

5. Objectives of the Proposed Research

The following objectives are determined
1. Development of Advanced Data mining classification models for human activities recognition.
2. Clustering models for grouping healthy and unhealthy persons based on their ECG signals.
3. Prediction models for determining the vital signs of persons based on the ECG signals during their activities.
4. Identification of physical human activities based on the mHealth sensor data sets.
5. Determination of basic health heart monitoring activities.
6. Determination of various arrhythmia or looking at the effects of exercise on the ECG values during physical
activities.
7. Determination of body’s basic function using ECG
8. Determination of person’s vital signs , which are varying with age, weight, gender, and overall health
9. Determination of the pulse rate (Heart Rate), body temperature, respiratory rate and blood pressure of the
normal and abnormal human being.

6. Research Methodology

DATA SCIENCE APPROACH FOR FOR HUMAN ACTIVITY RECOGNITION & VITAL SIGN
PREDICTTIONS

 The MHEALTH (Mobile HEALTH) dataset comprises Physical
activities and vital signs
 Recordings for ten volunteers of diverse profile while
performing several physical activities.
 Sensors placed on the subject's chest, right wrist and left
ankle are used to measure the motion experienced by
diverse body parts,
 namely, the acceleration, the rate of turn and the magnetic
field orientation.
 The sensor positioned on the chest also provides 2-lead ECG
measurements, which can be potentially used for basic heart
monitoring, checking for various arrhythmias or looking at
the effects of exercise on the ECG.

7.Research Plane with Time Schedule

Year Sl.
Nor
Progress task/per Every six month I Year
2015
II Year
2016
III Year
2017
1-6 7-12 1-6 7-12 1-6 7-12
I 1 Course Work Completion
2 Literature Survey on the proposed research
issue/Data collection/Research Toll Understanding
II 3 Data Model Design/Design and Development of
data processing Model
4 Development of Classification Model or HAR and
comparing with exiting Algorithms and
Evaluation(Model Deployment, Operations
and optimization)
III 5 Development of Predictive Model for Vital Sign
Analysis and their evaluation (Model
Deployment, Operation and Optimization)
6 Publication of Research outcome at National and
International Level journal/conference preceding
7 Preparation of Documentation of work for thesis
submission

8. Work Under Progress

Classification accuracies of various common exiting methods implemented on proposed mHealth Datasets (Banos et al,
2014) proposed for the prediction of vital signals of Human during their activities.
Sl nor Activity J48 N B REP TREE SMO Predictive Data Models
1 Walking 99.90 99.15 99.90 100 - -
SCOPE
OF EXPLORATION
.
.
.
??????
2 Jogging 97.23 93.33 97.94 95.46 - -
3 Running 96.19 92.74 91.86 91.33 - -
4 Knees Bending 100 99.86 100 100 - -
5 Lying down 100 100 100 100 - -
6 Jump F & B 93.18 93.18 97.77 91.61 - -
Further study is focused on the investigation of advance data mining algorithms and their evaluation on ECG Lead
data signals of mHealth data sets

The following activities need to predictive Vital sign
Climbing Stair
VITA SIGN PREDICTION FOR CLIMBING
STAIR

Running person Heart rate, BP, Temperature,
& respiration rate
Cycling Person HR, BP, Temperature, Respiration
rate
The Vital sign varies, which
are varying with age,
weight, gender, and
different physical activities
Running
Cycling Friday, August 12, 2016

According to mHealth dataset it contains 12 different physical activities and do 12 different vital function of all
activities
12 different physical from mHealth
Running person Heart
rate, BP, Temperature, &
respiration rate

Conclusions:
 Aim is to Develop classification methods for human activity recognition
and Vital prediction models.
 Literature survey is done on both human activity classification methods
and vital sign classification.
 Data Science approach is proposed for design , deployment , operation
of the proposed models.
 Primary data sets are obtained from mHealth Sensors.
 Plane of research is proposed.
 Little work carried out on the primary data sets using WEKA Tool.

Reference
2. Banos, O., Garcia, R., Holgado, J. A., Damas, M., Pomares, H., Rojas, I., Saez, A., Villalonga,
“mHealthDroid: a novel framework for agile development of mobile health applications”.
Proceedings of the 6th International Work-conference on Ambient Assisted Living an Active Ageing (IWAAL
2014), Belfast, Northern Ireland, December 2-5, (2014). Data set available at Download mHealth data sets.
http://www .northbridgeasia.com/research_reports.aspx

http://www.minoritynurse.com/article/

http://www.ugr.es/~oresti/datasets.htm

Thanks
Your suggestions and comments
Yogesh K M

Activity recognition in health field

More Related Content

Similar to Activity recognition in health field (20)

Recently uploaded (20)

Activity recognition in health field

Editor's Notes