Attrasoft DecisionMaker 2.5/Chap. 4

4. Application Database Examples

Figure 9. Examples.

All examples here are build in the DecisionMaker. To generate the Heart Disease Database, for example, click "Example/Heart Disease Database (13, 1), 5". Here "(13, 1), 5" means:

13 variables make up the questions
1 variable provides the answers
5 classes

Look at the status bar; it indicates this example has a total of 920 instances.

As a rule of thumb, your database should have 100 instances per attribute per class. If you increase the precision from 10 to 100, your database should have 200 instances per attribute per class. This amount of data can easily generate a prediction with 100% accuracy.

Example 1: 10 attributes, 2 classes.

100 * 10 * 2 ===> 2,000 instance for precision = 10.

200 * 10 * 2 ===> 2,000 instance for precision = 100.

If you do not have enough data, the accuracy will drop. If your data is substantially less than the above amount, you can only use the DecisionMaker (DM) for making an educated guess.

The databases used by the DecisionMaker as examples are:

Wisconsin Breast Cancer Database: Integer Weighted Average
Car Evaluation Database: Integer Distribution
Credit Approval Database: How to handle text Input
Ecoli database: Not enough data, educated guess
Boston Housing Database: continous classes.
Heart Disease Database: How to handle missing data
Forensic Science database: Not enough data, educated guess
Nursery School Database: enough data ==> 100% accuracy
Thyroid Database: 2 output variables

Depending on the amount of data you have, the DM 2.5 can be used in 2 ways:

Making a decision
Making an educated guess.

4.1 Wisconsin Breast Cancer Database

4.2 Car Evaluation Database

4.3 Credit Approval Database

4.4 Ecoli database: Protein Localization Sites

4.5 Heart Disease Databases

4.6 Boston Housing Database

4.7 Forensic Science Database

4.8 Nursery School Database

4.9 Thyroid Database

4. 10 More

Please see:

UCI Machine Learning Repository

At:

http://www.ics.uci.edu/~mlearn/MLSummary.html