Spatiotemporal Data Mining on Networks

1
Spatiotemporal Data Mining on Networks

• Taehyong Kim
• Computer Science and Engineering
• State University of New York at Buffalo

2
Table of Contents

• Studies
• Spreading and Defense model in Networks
• Fixed-random network
• Spreading Model
• Defense Model
• Avian Influenza Outbreaks
• Modeling
• Mining parameters

• Introduction
• Overview
• Networks Data Mining
• Spatiotemporal Data Mining
• Applications
• Quality of Bone (osteoporosis) as a Network
  Dynamics
• Amazon Deforestation

3
Overview

• Most of real world relationships and
  communications could be represented on networks
  (graphs).
• Understanding the behavior of such systems starts
  with understanding the topology of the
  corresponding network.

Yeast PPI network
ATT Web Network
Collaboration network
4
Overview

• Recent studies on various networks
• Social network
• Author network, School relationship Network
• Technical network
• Cell network, Internet, Electric power network
• Biological network
• Protein network, Metabolic network, Disease
  Network
• Focuses on network attributes
• Number of nodes and edges
• Weight on nodes and edges

5
Overview

• nodes and edges

Bridge node
edge
Hub node
node
6
Networks Data Mining

• Networks Data mining has been done
• Prediction of unknown protein functions in
  protein-protein interaction networks
• Resilience test of networks against attacks
• Prediction of people relationships in social
  networks
• Drug targeting on cell networks
• Etc.

7
Spatiotemporal Data Mining

• Networks are changed as time goes by
• World wide web is evolving by itself
• Interactions among proteins are changed in PPI
  networks
• Size of cities and inter-state free ways are
  changed
• Structure of bone is changed
• Information of location and time is also
  important factors for further understanding on
  any given networks

8
Spatiotemporal Data Mining

• Spatiotemporal Data Mining knowledge extraction
  from large spatiotemporal repositories in order
  to recognize behavioural trends and spatial
  patterns for prediction purposes
• What is the relationship between the spread of
  epidemics and the number and location of houses
  and schools by time?
• What is the connection between the size of
  Buffalo city and thruway traffics on I-90 by an
  year?

9
Spatiotemporal Data Mining
Normal
Osteoporosis
Drugs
10
Amazon Deforestation 2003
Deforestation 2002/2003
Deforestation until 2002
Fonte INPE PRODES Digital, 2004.
11
Amazon in 2015?
fonte Aguiar et al., 2004
12
Modelling Complex Problems

• Application of interdisciplinary knowledge to
  produce a model.

If (... ? ) then ...
Desforestation?
13
Table of Contents

• Studies
• Spreading and Defense model in Networks
• Fixed-random network
• Spreading Model
• Defense Model
• Avian Influenza Outbreaks
• Modeling
• Mining parameters

• Introduction
• Overview
• Networks Data Mining
• Spatiotemporal Data Mining
• Applications
• Quality of Bone (osteoporosis) as a Network
  Dynamics
• Amazon Deforestation

14
Spreading and Defense model in Networks

• Fixed-radius random network
• Cellular transmission tower
• Interstate free ways
• Epidemics on communities
• Sensor networks
• How we can defend if there are attacks or breaks
  from the center of the networks?

15
Fixed Radius Random Network

• 400 random points on 11 square unit
• Calculating distance between each point
• If two points are in a certain radius, creating
  an edge between points

16
Fixed Radius Random Network

• Fixed-radius of random network (r 0.01 0.14)

Fixed-Radius
400 nodes, 2366 edges
17
Simulation on network

• Network dynamics are studied based on
  fixed-radius random network
• Simple spreading model and defense model is
  implemented for simulation
• Mining important parameters on this model of
  network dynamics
• Mining optimal values of parameters on this model
  of network dynamics

18
Spreading Model

• Simulating disease spreading or message spreading
• Starting from center point (0.50.5)
• Affecting edges which are in a spreading radius
  (ROI) from center
• Spreading radius grows or reduces based on how
  many edges are damaged

19
Spreading Model

• Region of radial distance of spreading model
  (ROIt0 0.1)
• Spreading starts from center (0.5, 0.5)

ROI
Center
20
Spreading Model

• Probability of affecting rate of edges (Pa
  0.33)
• 11 edges are in ROI
• In this case, 4 out of 11 edges are affected
  (Spreading will affect edges about 33
  probability)

ROI
21
Defense Model

• Simulating defense system of disease spreading or
  message spreading
• Signaling to neighbor nodes in order to inform
  (disease) spreading
• Activated when the affection of spreading ( of
  signals from neighbor nodes) is over threshold
• Removing edges which are in a radius (f) from
  activated neighbor nodes in order to stop
  spreading

22
Defense Model

• Circular region of programming Cell Death
  (f0.23.6)
• When signals from neighbor nodes are over the Td,
  edges in the circular region are removed by
  defense process

Region of defense process
23
Defense Model

• Probability of Programming Cell Death (Pp 1)
• If Pp is 1, all edges in circular regions are
  dead

24
Result (visualization)

Time 0
Time 10
Time 50
Time ?
Total Damage

Intermediate
Contained




25
Result
26
Result
27
Summary

• Containment strategy on epidemics and virus
  spreads
• Mining important parameters
• Mining optimal values of important parameters
• Understanding dynamics on human tissues and bones
• Development of diseases (osteoporosis)
• Drug effects on cell networks

28
Table of Contents

• Studies
• Spreading and Defense model in Networks
• Fixed-random network
• Spreading Model
• Defense Model
• Avian Influenza Outbreaks
• Modeling
• Mining parameters

• Introduction
• Overview
• Networks Data Mining
• Spatiotemporal Data Mining
• Applications
• Quality of Bone (osteoporosis) as a Network
  Dynamics
• Amazon Deforestation

29
Avian Influenza

• AI outbreaks are frequently occurring around the
  world recently
• H5N1 type has high infection and mortality rate
• Chickens and ducks are main victims of AI
• Mortality rate of H5N1 could reach 90-100 within
  48 hours
• Threat from AI has greatly increased for human
  beings
• There are several reports showing human infection
  of AI
• People could get infected by contacting excretion
  of contaminated birds

30
AI outbreaks

• Outbreaks in South Korea 2008

31
AI outbreaks

• Outbreaks in South Korea 2008

4 days 12 days 20 days
28 days 36 days 44 days
32
Challenges

• Strategies are needed for AI containment
• Early identification of the first cluster of
  cases
• Warning system from contaminated area to neighbor
  areas are needed
• Effective quarantine plan should be existed
• Containment model helps plan effective strategies
• Prediction of damage with certain environment
  parameters
• Mining important parameters to control outbreaks
• Measurement of effective values of important
  parameters

33
Modeling

• A group of chickens and ducks are nodes
• 2231 nodes for a group of chickens and 808 nodes
  for a group of ducks
• 76 (1x1 square) units (1 unit 37.5 Km)
• Parameters
• A node can interact with other nodes in range g
• A susceptible node become a infected node by
  infection probability t
• A Infected node become a activated node by
  incubation period m and n
• Nodes are culled in quarantine radius l

34
Modeling
35
Visualization

• Visualization of simulations based on AI
  outbreaks in South Korea 2008

4 days 14 days 24 days
34 days 44 days
36
Important Parameters

• Effect of Increased Quarantine Range
• Quarantine radius 0.0 0.32 unit
• Effects of Increased Incubation Period
• Incubation Period 0 17 days
• Effects of Increasing the Infection probability
• Infection probability 0.0 1.0

37
Quarantine Radius

• Effect of Increased Quarantine Radius
• Quarantine radius 0.0 0.32 unit
• Infection probability 0.1, 0.4, 0.7 and 1.0
• Research on effective quarantine radius by
  Infection probability
• Optimal quarantine radius

Infection Probability 0.1 0.4 0.7 1.0
Optimal Radius 0.04 0.10 0.16 0.22
38
Quarantine Radius
39
Incubation Period

• Effects of Increased Incubation Period
• Incubation Period 0 17 days
• Quarantine Range 0.0, 0.04, 0.11 and 0.18 unit
• For mid level control, almost 89 of poultry
  farms are healthy when incubation period is one
  day whereas only 11 of poultry farms are healthy
  when incubation period is 17 days.

40
Infection probability

• Effects of Increasing the Infection probability
• Infection probability 0.0 1.0
• Quarantine Range 0.0, 0.04, 0.11 and 0.18 unit
• The large numbers of poultry farms eliminated by
  the aggressive culling procedure with max control

41
Summary

• Modeling AI dynamics based on statistic data
• Modeling of AI outbreaks and spreads
• Modeling of defense strategies
• Mining important parameters and values in order
  to contain AI outbreaks in early stage
• Quarantine radius, infection rate, incubation
  period
• Damage predictions with important parameters
• Mining defense strategies for future outbreaks

42
Thank you!