Transcript Helicoverpa Armigera

Understanding the Helicoverpa
Armigera Pest Population Dynamics
related to the chickpea crop using
Neural Networks
Rajat Gupta, B Narayana, Krishna Polepalli,
G. Ranga Rao, C Gowda, Y. Reddy and
G.Rama Murthy
INDEX







Introduction
Objective
Motivation
Pest Dynamics
Models Developed in the Past
Why they Failed ?
Preliminaries


Results



Dataset Description
Mean Graphs
Majority Voting
Conclusion
Introduction
Helicoverpa Armigera
Chickpea Crop
Participating Organizations
International Institute of
Information Technology
(IIIT)
International Crop Research
for the Semi-Arid Tropics
(ICRISAT)
Objective


To develop a pest forecasting mechanism
by extracting pest dynamics from Pest
surveillance database using Knowledge
Discovery and Data Mining techniques.
To understand the interaction of various
factors responsible for pest outbreaks.
Motivation


Insect pests are the major cause of crop
loss.
The crop loss due to lack of advance
information about pest emergence often
leads to financial bankruptcy of the
farmers.
Pest Dynamics

Highly dynamic nature of the Pest




Ability to adapt to new conditions quickly
Can migrate to long distances
Hibernate when condition are not favorable
Feeds on wide variety of hosts
Models Developed in the Past

Techniques used were essentially Statistical
(Correlation and Regression Analysis)

T.P. Trivedi had proposed a regression model to
predict the pest attack.


Model seems to work only for some years (1992-1994)
Correlation analysis was used by C.P. Srivastava to
explore the relationship between the rainfall and pest
abundance in different years.

The technique is not effective as the attributes don’t follow
normal distribution
Why they FAILED?



Techniques used are able to capture only
linear relationships.
Problems with the dataset (noisy data)
All events are treated equally
Pest Surveillance Dataset



Helicoverpa armigera pest data on
Chickpea crop provided by International
Institute for Semi-Arid Tropics (ICRISAT).
The dataset spans over a period of 11 years
(1991-2001).
It contains information on 17 attributes.
Dataset Description

These Dataset contains 17 attributes which
can be classified as



Weather parameters
Pest Incidence
Farm Parameters
Weather parameters





Rainfall
Relative Humidity
Minimum Temperature
Maximum Temperature
Sunshine hours.
Pest Incidence




Eggs/Plant
Larvae/Plant
Light Trap Catch
Pheromone Trap Catch
Farm Parameters






Zone
Location
Area Surveyed
Plant Protection
User
Season
Neural Networks



A Neural Network is an interconnected
assembly of simple processing elements,
units or nodes, called neurons.
The processing ability of the network is
stored in the inter-unit connection strengths
or weights.
Learns from a set of training patterns.
Multi Layer Neural Networks
Inputs
Outputs
Hidden Layer
Why Neural Networks ?


Neural Networks don’t make any distributional
assumption about the data.
It learns the patterns in the data, while statistical
techniques try to do model fitting.
This makes neural network modeling a
powerful tool for exploring complex,
nonlinear biological problems like pest
incidence.
Data Preprocessing




Data Selection
Data Reduction
Null Values
Data Transformation


Normalization
Fourier Transform
Neural Network Training

Dataset

Advance Dataset (X) where X =0,12,3.






Training Dataset - 8 years (1991 - 1998)
Test Dataset - 3 years (1998 - 2001)
Learning Algorithm – Levenberg-Marquardt.
Bayesian Regularization
Hyperbolic Tangent Sigmoid function in hidden
layers (2 hidden layers)
Linear Transfer function in outer layer
Datasets Generated




Advance (0)
Advance (1)
Advance (2)
Advance (3)
Average R-value
Dataset
Average R-value
(for 15 models)
Advance(0)
0.91
Advance(1)
0.96
Advance(2)
0.91
Advance(3)
0.75
Larvae/Plant -Advance(0)
Larvae/Plant -Advance(1)
Larvae/Plant -Advance(1)
Larvae/Plant -Advance(2)
Larvae/Plant -Advance(2)
Larvae/Plant -Advance(3)
Larvae/Plant -Advance(3)
Majority Voting
(40%)
# Hits
# Miss
# False Alarm
Advance(0) 27
4
6
Advance(1) 29
1
4
Advance(2) 27
2
12
Advance(3) 22
6
15
Majority Voting
(50%)
# Hits
# Miss
# False Alarm
Advance(0)
27
4
6
Advance(1)
28
2
2
Advance(2)
26
3
11
Advance(3)
22
6
12
Majority Voting
(60%)
# Hits
# Miss
Advance(0)
25
6
# False
Alarm
6
Advance(1)
26
4
2
Advance(2)
26
3
11
Advance(3)
21
5
12
Conclusion

We can now predict the pest attack using
Neural Networks two weeks in advance
with high probability.
References









Data Mining Concepts and Techniques By Jiawei Han and Micheline Kamber
Neural Networks A Comprehensive Foundation By Simon Haykin
Applied Multivariate Statistical Analysis By By Richard Arnold Johnson, Dean A. Wichern, Dean
W. Wichern.
Advanced Engineering Mathematics By Erwin Kreyzig.
Models for Pests and Disease Forecasting - T.P.Trivedi, D.K Das, A.Dhandapani and A.K.
Kanojia
Das D.K , Trivedi T.P and Srivastava C.P 2001. Simple rules to predict attack of Helicoverpa
armigera on crops growing in Andhra Pradesh, Indian Journal of Agricultural Sciences 71: 421423.
Zhongua Zhao, Zuorui Shen .Theories and their applications of Stochastic Simulation Models for
Insect population Dynamics. Department of Entomology, The China Agricultural University.
``http://www.cau.edu.cn/ipmist/chinese/lwzy/xuweilw/xwlw-zhzhao.htm``
Agarwal, R., Imielinshki, T., Swami, A. 1993. "Mining association rules between sets of items in
large databases" .Proc. of ACM-SIGMOD Int'l Conf. on Management of Data: 207-216.
Agarwal, R. Srikant, R., 1994, Fast Algorithms for Mining Association Rules, Proc. of the 20th
VLDB: 487-499.