## What does it mean *generalization parameter *?

In this realization you can choose generalization parameter value between 0.1 and 100. Default value is 1 and it is suggested as your first choice.

High parameter value means that you want models that cover many positive examples even if this result is paid so that the model incorrectly classifies some negative examples as positive. In other words, high generalization parameter values mean that the induced models will satisfy more examples and that among them will be many positive examples but also some negative examples which should not be there. High generalization parameter value is suggested for problems where incorrectly classified examples can be expected in the input data file (noise) as well as for problems in which there is no strong connection between the target attribute and input attribute values.

In every inductive problem it is suggested to make few experiments with different generalization parameter values. It can not be expected that for every parameter value different models will be induced. Also, the same generalization parameter value is not necessary appropriate for different problems. Problems with many examples in the data file typically require higher parameter values. Expert understanding of the problem may significantly help in selection of the most appropriate value.

Default value of the generalization parameter equal

**1**will typically induce models that are very specific. Their covering of positive examples can be expected to be rather modest. Parameter values

**below 1**will seldom give new useful models while levels

**greater than 10**may generate very general models describing global domain properties.

© 2001 LIS - Rudjer Boskovic Institute

Last modified: April 23 2017 05:58:29.