## What does it mean *number of models *?

For your data there is a possibility to construct up to three models. Their number is selected by this option.

There exist the possibility that the

**number of actually presented models is less**than the requested number. The reason is that some of the induced models may be identical and copies of the same model are not presented.

The

**first model**is always constructed in the same way, regardless if second and third models will be constructed. The selection of the first model is based on the covering properties over the complete set of available examples. Selected is the model with maximum ILLM covering quality. It must be noted that computed quality, except on the number of covered examples, depends also on the generalization level selected by the user.

After first model selection, relative weight of every positive example covered by the first model is decreased. Selection of the

**second model**is based on the same ILLM covering weight formula but now preference is given to models which cover positive examples that were not covered by the first model. In this way it is expected that second selected model will be different from the first one. But in situations when the first model is significantly better than all other models, then it can happen that the second selected model is the same as the first one.

After second model selection, the same procedure is repeated for the

**third model**. At first, the weight of positive examples covered by the second model is decreased. It means that examples covered both by the first and the second model will enter the third iteration with very low relative weight. The third model will be equal to the first or to the second selected model only in situations when there is no good model covering already uncovered positive cases.

© 2001 LIS - Rudjer Boskovic Institute

Last modified: September 09 2015 14:17:42.