|
The explosive growth in our capabilities to generate and
collect data has made the field of knowledge discovery in
database emerging. Most of the corporate database management
systems provide managers with query tools that their utilization
is usually limited by the complexility of business decision
problems. In addition, the volumes of data and the unknown
relationships between decision variables both make the decision
process even hard. This research investigates the usage of
neural networks and conditional probabilities in discovery
information which can support the business decision process from
the corporate database. To evaluate the method, the
decision problem of underwriting of insurance policy is
examined. The underwriting decision is supported with the use of
neural networks with back propagation learning algorithm.
Besides, some data rules is found by the computation of
conditional probabilities. By combining these rules with the
neural networks, better networks are expected. The
results of the experiments show that the configuration of
network structure affects the performance of neural networks. In
predicting the underwriting decision, the best error rate is
below 29%. The performance of networks that combined with data
rules is not significant different from those of other networks.
The research also builds another set of neural networks on the
results of the sensitivity analysis of one neural networks. The
average performance of these set of neural networks is much
better. The sensitivity analysis provides a way to understand
the knowledge discovered by the neural networks.
|