Neural Networks in the Prediction of Avian Risk Status

1
Neural Networks in the Prediction of Avian Risk
Status

• Cole Flohr
• April 27, 2006

2
Problem

• Develop a network to predict the degree of
  extinction risk for bird species based on life
  history data

3
IUCN Criteria

• Rate of decline
• Population size
• Area of geographic distribution
• Degree of population and distribution
  fragmentation

• Extinct
• Extinct in the Wild
• Critically Endangered
• Endangered
• Vulnerable
• Near Threatened
• Least Concern
• Data Deficient

4
Goals

• Develop a risk prediction network that does not
  rely on measuring population declines
• Habitat decline
• Possible relationships to life histories
• Investigate which factors seem to be most
  indicative of extinction risk

5
Network Design

• Target Function
• IUCN Red List Category
• Least Concern or At Risk
• Data North American Birds
• 2,000 Birds, non-Hawaian
• 600-700 available data sets
• Nesting, Breeding, and Feeding Habits
• Source Ehrlich, Dobkin, Wheye

6
Nesting Data

• Nest Location
• ( hight from ground)
• Floating
• Ground
• Shrub
• Tree
• Cliff

• Nest Type
• Scraped
• Cup
• Platform
• Pendant
• Cavity
• Burrow

7
Mating and Rearing Roles

• Mating System
• Monogamy
• Polygyny (M F)
• Polyandry (F M)
• Promiscuity (indescriminate)
• Polygamy (polygyny and polyandry)
• Cooperative (Females rear broods together,
  non-breeding birds help parents)

• Gender Roles
• Nest Builder
• Incubator
• Tending Young

8
Development

• Development at Hatching
• Precocial
• Independent, eyes open, downy
• Semiprecocial
• Dependant, eyes open, downy
• Altricial
• Helpless, blind, featherless

• Development Rates
• Incubation time
• Hatching to Fledging (in Days)

9
Behavior and Diet

• Primary Diet
• Greens
• Insects
• Fish
• Vertebrates
• Fruits
• Nectar
• Nuts
• Omnivore

• Foraging Technique
• Glean (Ground, Foliage, bark, hover)
• Hawks
• Probes
• Aerial Foraging
• Dabbles
• Stalk and strike
• Hover and Pounce
• Surface Dives/Dips

10
Numerical Input
Unknowns substituted with average values for
similar species
11
Network Construction

• SNNS (JavaNNS)
• 15 input nodes
• 8 or 15 hidden nodes in one layer
• 6 or 2 output nodes

12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Sample Results

• 7.1
• 10 0.1 1 3 12 16 0.1 3 23.5 11
• 6.5 3 25 32 1
• 0 1
• 0.03706 0.96277
• 10.1
• 10 0.1 7 0 1 1 0.1 3 44 1
• 60 3 4 6 22
• 0 1
• 0.09207 0.90872
• 11.1
• 15 15 4 3 1 1 0.1 3 50 0.1
• 170 3 4 6 20
• 1 0
• 0.99881 0.00121
• 12.1
• 10 0.1 4 3 2 4 0.1 4 21 2
• 28 3 6 5 34
• 1 0

• SNNS result file V1.4-3D
• generated at Tue Apr 25 193104 2006
• No. of input units 15
• No. of output units 2
• 1.1
• 10 0.1 1 2 4 7 0.1 2 22.5 15
• 42.5 3 25 6 31
• 1 0
• 0.99305 0.00692
• 2.1
• 10 0.1 1 2 4 7 0.1 2 27.5 15
• 56.5 3 25 6 31
• 1 0
• 0.99312 0.00686
• 3.1
• 10 0.1 1 2 8 10 0.1 2 27.5 15
• 61.5 2 6 32 32
• 1 0
• 0.97757 0.02237

20
(No Transcript)
21
(No Transcript)
22
Analysis-New Examples

• 32.1
• 15 4 2 3 3 4 0.1 3 14 0.1
• 13 3 5 21 1
• 0 1
• 0.09803 0.90208
• 33.1
• 10 0 2 2 4 5 0.1 2 10 0.1
• 9.5 3 5 -1 2
• 0 1
• 0.86904 0.12982

• 34.1
• 10 0 2 2 4 5 0.1 4 11.5 0.1
• 11 4 5 -1 2
• 1 0
• 0.88426 0.11459
• 35.1
• 20 15 8 2 4 5 0.1 4 13 0.1
• 10 4 5 -1 2
• 1 0
• 0.80259 0.19548

23
Preliminary Conclusions

• Network can learn risk status from life history
  data
• Accuracy on training set gt 95
• Accuracy on new examples
• 50 percent for 6 output nodes
• 70 percent for 2 output nodes

24
Ongoing work

• Continue error analysis
• Use the six-category target function with a
  single output node
• Retrain networks with larger example set
• Compare results from a Passeriformes only network
  to all families network
• Investigate the contribution of certain categories

25
Questions?
26
Weights

• Nest Location
• Nest Height
• Nest Type
• Nest Builder
• Clutch Min
• Clutch Max
• Mating System
• Incubating Sex
• Incubation Time
• Dev. at Hatching
• Time to Fledge
• Tending Parent
• Diet 1
• Diet 2
• Foraging Technique