X 2 . Again, how well did this work? Plot your

b 2 X 2 . Again, how well did this work? Plot your

results. Is anything wrong? Calculate a sum of the

squares error and R 2 .

d. Break up the problem into three sections (look at

the plot) and solve it with three linear regression

61.9 models, one for each section. How well did this

work? Plot your results. Calculate a sum of the

squares error and R . Is this modeling approach

appropriate? Why or why not?

e. Build a neural network to solve the original prob- lem. (You may have to scale the X and Y values to

. C H A P T E R 1 2 A D V A N C E D INTELLIGENT SYSTEMS

each of the original data items). H o w well did this or even survey the class for weight, gender, and work? Plot your results. Calculate a sum of the

height and try to predict height based on the other

two. (Hint: U s e U.S. Census data, on this book's Web f. Which method worked best and why?

squares error and R 2 .

site or at www.census.gov, by state to identify a rela- 3. Build a real-world neural network. Using demo soft-

tionship between education level and income.) H o w

ware downloaded from the Web (Braincel at

good are your predictions? Compare the results to www.promland.com or another), identify real-world

predictions generated by standard statistical methods

data (e.g., start searching on the Web at

(regression). Which method was better? H o w could www.research.ed.asu.edu or use data from an organi-

your system be embedded in a DSS for real decision zation with which someone in your group has a con-

making?

tact) and build a neural network to make predictions. 4. Fuzzy logic. Survey your class by having everyone Topics might include sales forecasts, predicting suc-

write down a height representing tall, medium, and cess in an academic program (predict GPA from high

short for men and for women. Tally the results and school rating and SAT scores—see ftp://psych

determine what is meant by tall, medium, and short .colorado.edu/pub/stat/gpa.txt, being careful to look

in a fuzzy way. Create the membership functions in out for "bad" data, e.g., GPAs of 0.0), housing prices,

these sets and examine the results.

• INTERNET EXERCISES 1. Case-based reasoning has been used lately for data

5. Examine fuzzy logic vendor Web sites and identify mining. Explore the Web to find vendors and

the kinds of problems to which fuzzy logic is cur- research literature about this topic,

rently being applied. Find a demo version of a 2. Explore the Web sites of several neural network

system and try it out. Report your findings to the vendors, such as California Scientific Software

class.

(www.calsci.com), NeuralWare Inc. (www.neuralware 6. U s e the Internet to find information about neuro- .com), and Ward Systems Group (www.wardsystems

fuzzy logic systems.

.com), and review some of their products. Download 7. Access the Web and e-journal in your library to at least two demos and install, run, and compare

find at least three reports on the use of integrated them.

methods for intelligent decision support. Evaluate 3. Explore the Web to identify the current status of

whether the applications are feasible in the real neural network research.

world.

4. Examine genetic algorithm vendor Web sites and investigate their business applications. What kinds of applications are most prevalent?