Artificial datasets

Now to the first tests. There are several reasons why artificial datasets are useful. These include:

• Evaluation is simplified. For example, the induced concept can be compared with the correct concept. It can also be used to explore factors of interest such as noise, etc.
• There is a chicken-and-egg problem as far as temporal learners and datasets are concerned. Because there are few (if any) generic temporal learners, people do ad hoc feature extraction themselves. For example, an examination of the UCI repository ([MM98]) reveals at least seven datasets which are inherently temporal in nature, but where the researchers did the feature extraction themselves and the raw data is not available.

• In order to compare our results with other researchers.