Transcript Reverse Engineering XML Schema to UML

Mobile Data Mining for Intelligent
Healthcare Support
By: Pari Delir Haghighi, Arkady Zaslavsky, Shonali Krishnaswamy,
Mohamed Medhat Gaber
Center for Distributed Systems and Software Engineering
Monash University, Australia
www.monash.edu.au
An Overview
• Introduction
• The State-of-the-Art
• Situation-Aware Adaptive Processing (SAAP) of
Data Streams
• Fuzzy Situation Inference (FSI)
• Adaptation Engine (AE)
• Implementation
• Evaluation
• Future Work
• Conclusion
www.monash.edu.au
2
Introduction
Mobile healthcare services:
• provide a convenient, safe and constant way of
monitoring of vital signs
• development of mobile healthcare applications
encouraged by
– innovations in mobile communications
– low-cost of wireless biosensors
• the issues:
– maintaining continuity of running applications on mobile
devices
– enabling real-time analysis of data and decision making
www.monash.edu.au
3
The State-of-the-Art (1)
• recent works in mobile healthcare
– mostly focused on using, enhancing or combining existing
technologies
> projects: EPI-MEDICS [RFN05],MobiHealth [MWH07]
– limited use of context-awareness
– lack of resource-aware data analysis techniques
• a need for a general approach:
– performing smart and cost-efficient analysis of data
in real-time
– providing a general model for representation of
real-world and health-related situations
www.monash.edu.au
4
The State-of-the-Art (2)
Ubiquitous Data Stream Mining (UDM)
– real-time analysis of data streams on-board
small/mobile devices
> techniques and algorithms for resource-aware data
stream mining [GKZ05]
• However, to perform smart and intelligent
analysis of data on mobile devices
– imperative to factor in contextual information
www.monash.edu.au
5
Situation-aware Adaptive Processing
(SAAP) of Data Streams
SAAP:
1. incorporates situation-awareness into data
stream mining
2. performing situation-aware adaptation of data
streaming parameters according to occurring
situations and available resources
3. situation-awareness achieved by Fuzzy Situation
Inference (FSI) model
– FSI combines fuzzy logic principles with the
Context Spaces (CS) model
>
a general context modeling and reasoning approach for
pervasive computing environments
www.monash.edu.au
6
The Framework of SAAP
www.monash.edu.au
7
Fuzzy Situation Inference (FSI)
• FSI inspired by the Context Spaces (CS) Model [PAD04]
• The CS model
advantages:
> deals with uncertainty associated with sensors’
inaccuracies
disadvantages:
> does not deal with other aspect of uncertainty related to
human concepts and real-world situations
• FSI integrates fuzzy logic principles into the CS model FSI
– enables representation of vague situations
– reflects minor and delta changes in the inference results
www.monash.edu.au
9
FSI: Situation Modeling
• linguistic variables: e.g. heart rate
• terms/Fuzzy sets: e.g. low, normal, fast
• membership functions to map input data into fuzzy sets
• A FSI Rule defines a situation
– consists of multiple conditions joined with the AND operator
> each condition can be a disjunction of conditions
e.g. if Room-Temperature is ‘hot’ and Heart-Rate is ‘fast’ and (
Age is ‘middle-aged’ or ‘old) then situation is ’heat stroke’ ’
www.monash.edu.au
10
Reasoning Techniques (1,2, 3)
Reasoning technique 1
Heuristics: weight and contribution
CS
Reasoning technique 2
Heuristics: sensors’ inaccuracy
n
Confidence   wi ci
CS
i 1
FSI
n
Confidence   wi  ( xi )
FSI
i 1
n
Confidence   wi . Pr( aˆit  Ai )
i 1
n
Confidence   wi  ( f ( xi , ei ))
i 1
Reasoning technique 3 and 4
Heuristics: Symmetric and Asymmetric context attributes, partial and complete containment
CS
n
Confidence   wˆ i . Pr( aˆ it  Ai )
i 1
FSI
where
n
Confidence   wˆ i  ( f ( xi , ei ))
where
i 1
CS
FSI
n
ai  CAS  CAA
xi  FS and FS  LVS  LVA
m
where
k 1
m
and
Confidence  q1  wˆ i . p(aˆ  Ai )  q2  p(aˆ kt  Ak )
i 1
n
t
i
q1  q2  1
ai  CAS  CAA , ak  CAS
Confidence  q1  wˆ i . ( f ( xi , ei )  q2   ( f ( xk , ek ))
i 1
k 1
www.monash.edu.au
11
SAAP
• Fuzzy Situation Inference (FSI) Engine
• Adaptation Engine (AE)
– Resource-aware strategies
– Situation-aware strategies
– Hybrid strategies
www.monash.edu.au
12
Adaptation Engine (AE)
www.monash.edu.au
13
The Controller
Cases
Adaptation Strategy
1 – R at safe level and S at safe Level
Situation-aware
2 – R at safe level and S at medium level
Situation-aware
3 – R at safe level and S at critical level
Situation-aware
4 – R at medium level and S at safe level
Resource-aware
5 – R at medium level and S at medium level
Hybrid
6 – R at medium level and S at critical level
Hybrid
7 – R at critical level and S at safe level
8 – R at critical level and S at medium level
9 – R at critical level and S at critical level
Other strategies e.g.
migration
www.monash.edu.au
14
Resource-aware Adaptation
• Lightweight data stream mining algorithms
– Adjusting mining parameters according to resource
availability
– E.g: LWC (LightWeight Clustering) [GKZ05]
> considers a threshold distance measure for clustering
> Increasing the threshold discourages forming of new
clusters
– in turn reduces memory consumption
www.monash.edu.au
15
Situation-aware Adaptation
• based on the concept of resource-aware
adaptation
• but adjustment of parameters according to
results of situation inference (FSI engine)
• starts with pre-set values of parameters for
each situation
• at run-time based on degree of fuzziness of
each situation these parameters adjusted
n
n
i 1
i 1
pˆ j   i p j /   i
µ: degree of fuzziness of each situation
p: parameter value
www.monash.edu.au
16
Hybrid Adaptation
• when both resources and situations are getting
critical
• a trade-off between the results of these two
strategies
• hybrid method combines resource-aware and
situation-aware strategies and deals with the
trade-off:
criticality of resources and
( pˆ R .criticalit y R )  ( pˆ S .criticalit y S ) situations represented by a
pˆ I 
value between 0 and 1
criticalit y R  criticalit y S
www.monash.edu.au
17
Implementation
• healthcare monitoring application
• Implemented in J2ME
• deployed on a Nokia N95 mobile
phone
• situations: ‘normal’, ‘PreHypotension’, ‘Hypotension’,
‘Hypertension’ and ‘Pre-Hypertension’
• context: SBP, DBP and HR
• using a Bluetooth-enabled ECG
sensor
www.monash.edu.au
18
Evaluation of FSI
A Comparative Evaluation
• The reasoning approaches
– FSI
– CS
– Dempster-Shafer (DS)
• to highlight the benefits of the FSI for reasoning
about uncertain situations
www.monash.edu.au
19
FSI Evaluation: Dataset
•
The dataset:
– generated continuously (data rate is 30 records/minute) in ascending
order
– 131 context states
– used our data synthesizer
> to represent the different events (of the DS model)
– contribute to the occurrence of each pre-defined situation as well as the
uncertain situations
Context attribute scales
SBP:40-65, DBP: 20-45, HR: 20-45
SBP:66-80, DBP: 46-60, HR: 46-60
SBP:81-85, DBP: 61-65, HR: 61-65
SBP:86-105, DBP: 66-85, HR: 66-85
SBP:106-130, DBP: 86-110, HR: 86-110
SBP:131-135, DBP: 111-115, HR: 111-115
SBP:136-170, DBP: 116-150, HR: 116-150
Corresponding DS events
SBPLow, DBPLow, HRSlow
SBPLow, DBPLow, HRMed
SBPLow, DBPMed, HRMed
SBPMed, DBPMed, HRMed
SBPMed, DBPMed, HRHigh
SBPLow, DBPHigh, HRHigh
SBPHigh, DBPHigh, HRHigh
www.monash.edu.au
20
FSI Evaluation: Results
Level of Confidence
Comparison of DS, CS and FSI for Hypertension
1.2
1
0.8
FS_Hyper
0.6
CS_Hyper
0.4
DS_Hyper
0.2
0
1
11
21
31 41
51
61
71
81
91 101 111 121 131
Comparison of DS, CS and FSI Hypotension
Data Row s
Level of Confidence
1.2
1
0.8
FS_Hypo
0.6
CS_Hypo
0.4
DS_Hypo
0.2
0
1
Comparison of DS, CS and FSI for Normal
Level of Confidence
1
0.8
FSI_N
0.6
CS_N
0.4
DS_N
0
1
11
21
31
41
51
61
71
Data Row s
81
91 101 111 121 131
21
31
41
51
61
71
Data Row s
1.2
0.2
11
81
91 101 111 121 131
FSI Evaluation: Results
• when situations are stable and pre-defined (not vague)
– all have a relatively similar trend
– more noticeable with the CS and FSI models
• when situations change and evolve
– the CS and DS methods show sudden rises and falls with
sharp edges
> not matching the real-life situations
– Yet FSI reflects very minor changes between situations
> represent changes in a more gradual and smooth manner
> more appropriate approach for health monitoring
applications
www.monash.edu.au
22
Evaluation of Situation-aware Adaptation
• Data stream mining algorithm used
– the LWC algorithm
• situations
– ‘normal’, ‘hypertension’ and ‘hypotension’
– situations’ importance: 0.1, 0.9 and 0.5
– parameter set values: 42 (normal), 10 (hypertension) and 26
(hypotension)
– context attributes: SBP, DBP and HR
• Dataset
– the same used in the FSI evaluation
> 131 context states (rows)
www.monash.edu.au
23
SA Evaluation: Results
Level of Confidence of Situation
1.2
1
0.8
FSI_N
0.6
FS_Hypo
FS_Hyper
0.4
0.2
0
26
26
26
29
32
42
42
35
35
29
10
10
10
10
Data Stream Algorithm Threshold
www.monash.edu.au
24
SA Evaluation: Results
• threshold value automatically adjusted according
to the fuzziness and membership degree of each
situation
• when situations are normal, threshold increases
– increasing the threshold value for normal situations
decreases the mining output
– reduces resource consumption
• when situation get critical, threshold decreases
– increases the number of the output (clusters) and
accuracy level of results that is required for closer
monitoring
www.monash.edu.au
25
Future work
• currently finalizing implementation and
evaluation of hybrid adaptation using RACluster
• using RA-Cluster enables adaptation of
sampling rate according to battery charge
• integrating time-constraint into adaptation of
battery usage
• working on testing of our prototype in realworld situation in conjunction with relevant
healthcare professionals
www.monash.edu.au
26
References
[GZK04] Gaber MM, Zaslavsky A, Krishnaswamy S (2004), A Cost-Efficient Model for
Ubiquitous Data Stream Mining, Proceedings of the Tenth International
Conference on Information Processing and Management of Uncertainty in
Knowledge-Based Systems, Perugia Italy.
[GKZ05]Gaber MM, Krishnaswamy S, Zaslavsky A (2005) On-board Mining of Data
Streams in Sensor Networks”, A Book Chapter in Advanced Methods of
Knowledge Discovery from Complex Data, (Eds.) S. Badhyopadhyay, U. Maulik,
L. Holder and D. Cook, Springer Ver-lag.
[MWH07] Mei, H., Widya, I., Halteren, A.V., and Erfianto, B., A Flexible Vital Sign
Representation Framework for Mobile Healthcare. 2007.
[PLZ05] Padovitz, A., Loke, S.W., Zaslavsky, A., Burg, B. and Bartolini, C.: An
Approach to Data Fusion for Context-Awareness. Fifth International Conference
on Modeling and Using Context, CONTEXT’05, Paris, France (2005).
[PZL06] Padovitz, A., Zaslavsky, A. and Loke, S.W.:. A Unifying Model for
Representing and Reasoning About Context under Uncertainty, 11th
International Conference on Information Processing and Management of
Uncertainty in Knowledge-Based Systems (IPMU), July 2006, Paris, France
(2006).
[RFN05] Rubel, P., Fayn, J., Nollo, G., Assanelli, D., Li, B., Restier, L., Adami, S., Arod,
S.,Atoui, H., Ohlsson, M., Simon-Chautemps, L., Te´lisson, D., Malossi, C., Ziliani,
G., Galassi, A., Edenbrandt, L., and Chevalier, Ph., Toward Personal eHealth in
Cardiology: Results from the EPI-MEDICS Telemedicine Project. Journal of
Electrocardiology 2005. 38: p. 100-106
www.monash.edu.au
27
Thank you
Questions?
www.monash.edu.au
28