Practical Applications of Data Mining

86 Chapter 3 ClassifiCation learning

After the accuracy of the rule reaches 100% or there are no more 3.

attributes left, the process of adding additional tests to the condition term

of the rule stops, and the positive instances covered by the rule are re-

moved from the dataset.

The Prism algorithm is summarized in Figure 3.2 below:

In real-world applications, data is not always as sound and clean as in our

example, especially in large databases. In these circumstances, the 100% accu-

racy requirement of the rule cannot be satisfied. The Prism algorithm cannot

handle this situation and potentially useful information might be ignored. In

some applications, a certain level of error is allowed for predictions.

To overcome the over-fitting ...