# Using Neural Networks and OLAP Tools to Make Business Decisions

In the Detailed Report to the right of the data set, Auto-Prediction provides the predicted outcome for the new applicants, along with the probabilities of those outcomes. With Live Prediction (discussed in Appendix 3) enabled, the prediction and the probability change as independent values change. For example, with a lower loan amount, the predicted probability of this applicant falling within the "timely payments" classification typically will be higher. The Detailed Report to the right of the data set shows not only the probability of the predicted category, but also the probability of every possible classification for each applicant. Those can be obtained by making appropriate selections in the Application Setting dialog: by clicking "Columns in Detailed Reports" and in the table that shows selecting "Probabilities of All Categories (for PNN)" in the Prediction column. Refer to Figure 5.

By using another Excel add-in from Palisade, Evolver, with NeuralTools, you can solve optimization problems, as seen in the Evolver model definition shown in Figure 6. A bank employee wants to allocate $75,000 in loans to the five applicants so as to minimize the probability of a default occurring. That probability is calculated by the formula in cell P22. The Evolver model has been set up as follows: The optimization goal was defined by specifying cell P22 and selecting "Minimum" as Optimization Goal. Range F26:F30 was specified as the adjustable range, with $0 and $30,000 as minimum and maximum values. Also, the Budget solving method was selected (by clicking Group button, selecting Edit, and selecting Budget), to ensure the total of cells F26:F30 will be kept constant. To run the example, select Evolver Start Optimization icon shown at the top right in Figure 4.

*Click here for a larger image.*

**Figure 6:** A bank wants to allocate $75,000 in loans to five applicants so as to minimize the probability of a default occurring.

*Click here for a larger image.*

**Figure 7:** Decision making tools discussed in this and the follow-uparticle (*NeuralTools with Evolver* and *NeuralTools with StatTools*, respectively).

Here's another optimization that can be solved by NeuralTools and Evolver applied to the very same data set used in the previous example:

A lending institution can determine how much to loan a new applicant. The bank wants a repayment probability of 90%. Less than 90% is too risky, and over 90% means the bank is being too cautious and not lending enough. The amount to loan the applicant is an adjustable cell (K28 in Figure 8). Evolver will try different loan amounts and NeuralTools will predict the probability of repayment live, during optimization, while Evolver will determine the best loan amount to achieve 90% repayment.

*Click here for a larger image.*

**Figure 8:** A bank wants a repayment probability of 90%.

The Appendices and References sections provide further information on the NeuralTools *types of net* and the Evolver *solving method* (Budget and Recipe) used above.

Note:The example given above requires Evolver and the Industrial version of NeuralTools (to support Live Prediction).

### Conclusion

OLAP and neural network users frequently have different characteristics. Those working with OLAP may employ software to access predefined reports, manipulate the data using available dimensions and measures and (in the case of power users) create queries and reports for themselves and others.

Neural network analysts typically work with specialized software to find the relationships that are important to the business. These analysts may be either highly skilled professionals or businesspeople with good analytical and problem-solving skills who work with packaged software in applications such as fraud detection. Of course, these analysts and the work they do can differ considerably.

A recent article in the *Boston Globe* newspaper discussed a case of the improper deployment of the kind of software discussed above. The Commonwealth of Massachusetts had planned to distribute sophisticated management software to 20,000 state government users, many of whom, according to industry analysts, would lack either the skills or the need to use it.

Page 3 of 6