Neural Network Classification and its Applications in Insurance Industry
1. Inderjeet Singh
7667292
Department of Computer Science
University of Manitoba
2. Introduction
Classification with Neural Networks
a) Advantages
b) Disadvantages
Performance Studies
Application (Insurance Industry)
Conclusion
References
3. Neural Networks are models of intelligence
that consist of large numbers of simple
processing units that collectively are able to
perform very complex pattern matching
tasks. These models perform stimulus
response mapping
Classification is the process of learning rules
or models from training data to generalize
the known structure and then to classify new
data with these rules
4. Advantages (motivations)
1. Data driven and self-adaptive
2. Universal function approximators
3. Non-linear model making, flexible for real
world applications
4. High accuracy and noise tolerance
5. Disadvantages (problems)
1. Lack of transparency (black box)
2. Learning time is long (trail and error)
3. Defining classification rules (rule extraction)
is difficult
6. Comparison of Neural classifier [Lu et al.] and
decision tree classifier
People database consisting of 9 attributes
(age, elevel, zipcode .etc.) and 1 output (Group A
or Group B)
3 layer feed forward neural network (38 input
units, 6 hidden units and 1 output unit)
Tested and compared their approach on 8
classification problems used in earlier researches
Func 3
7. Accuracy of rules extracted from The number of rules extracted
neural networks (NN) and C4.5 from neural networks (NN) and
algorithm (DT) C4.5 algorithm (DT)
8. The number of conditions per
neural network rule (NN) and
C4.5 rule (DT)
10. Profit and growth
Neural networks: Understanding customer
retention patterns (renewal or termination)
Helps in Predicting likely terminations
Direct marketing campaigns
Misclassification costs
Accuracy is important
Helps in Price setting (balanced profit and
growth)
12. 3 layer feed forward neural network, with
hyperbolic tangent activation function and
conjugate gradient technique to minimize
the error
29 input nodes (attributes), 25 hidden
nodes and 1 output node, dataset-20914
motor vehicle policy holders
Neural classifier outperformed regression
analysis and decision trees
13. Lift Chart: Percentage of policy holders
classified for likely termination vs
Percentage of policy holders selected from
the test dataset
14. Scope of improvement in terms of speed of
classification
Suits the need of many business applications
which have lots of data available
15. 1. Hongjun Lu, Rudy Setiono and, Huan Liu, Effective Data Mining
Using Neural Networks, Vol 8, IEEE Transactions on Knowledge
and Data Engineering,1996, pp. 957-961
2. David Scuse, Chapter 1 Intro, Class slides, University of
Manitoba
3. Wikipedia.com: http://en.wikipedia.org/wiki/Data_mining
4. K.A. Smith, R.J. Willis and M. Brooks, An Analysis of Customer
Retention and Insurance Claim Patterns Using Data Mining: A
Case Study, The Journal of the Operational Research
Society, Vol. 51, May 2000, pp. 532-541
5. Image: http://www.genevievecharest.com/2011/09/26/do-a-
easy-vehicle-insurance-comparability-before-choosing-an-
auto
Hinweis der Redaktion
conclusion
Learning takes a lot of passes over the training data, so training time is long (trail and error)Defining classification rules (rule extraction) is difficult due to the complex structure of network and weights learned by branches between nodes