Page 201 - DCAP208_Management Support Systems

Management Support Systems

Notes 12.1 Applications of ANN

Neural networks have been successfully trained to determine whether loan applications should
be approved. It has also been shown that neural networks can predict mortgage applicant
solvency better than mortgage writers. Predicting rating of corporate bonds and attempting to
predict their profitability is another area where neural networks have been successfully applied.
Neural networks outperformed regression analysis and other mathematical modeling tools in
predicting bond rating and profitability. The main conclusion reached was that neural networks
provided a more general framework for connecting financial information of a firm to the
respective bond rating.
Fraud prevention is another area of neural network application in business. Chase Manhattan
Bank successfully used neural networks in dealing with credit card fraud (Rochester, 1990), with
the neural network models outperforming traditional regression approaches. Also, neural
networks have been used in the validation of bank signatures. These networks identified forgeries
significantly better than any human expert. Another significant area of statistical application of
neural networks is in time series forecasting. Several studies have attempted to use neural
networks for time-series prediction.
Examples include Fozzard et al. (1989), Tang et al. (1991), and Hill et al. (1994).
The general conclusion is that neural networks appear to do at least as well as the Box-Jenkins
forecasting technique.
Because neural networks have been a subject of intense study since late 1980s, there have been
many applications as well as experiments with applications. Other recent reports include live
intrusion tracking , Web content filtering, exchange rate prediction, and hospital bed allocation.
Newer applications are emerging in health care and medicine.
In general, ANN are suitable for problems whose inputs are both categorical and numeric, and
where the relationships between inputs and outputs are not linear or the input data are not
normally distributed. In such cases, classical statistical methods may not be reliable enough.
Because ANN do not make any assumptions about the data distribution, their power is less
affected than traditional statistical methods when data are not properly distributed. Finally,
there are cases in which the neural networks simply provide one more way of building a
predictive model for the situation at hand.

Did u know? Given the ease of experimentation using the available software tools, it is
certainly worth exploring the power of neural networks in any data modeling situation.

Self Assessment

State True or False:

1. Neural networks outperformed regression analysis and other mathematical modeling
tools in predicting bond rating and profitability.
2. Neural networks are not used in dealing with credit card fraud .

3. We can use neural networks for time-series prediction.
4. ANN are not suitable for problems whose inputs are both categorical and numeric.

12.2 Other Applications

In this section, we will discuss various applications of ANN.

194 LOVELY PROFESSIONAL UNIVERSITY