ECIS can:

• Look for the best representation in the sense of different criteria, like predictive classification accuracy, complexity of description, cost etc.
• Introduce new attributes, select the existing ones, discard the redundant/irrelevant ones
• Be configured according to the user's wishes: probabilities of operators, relations used etc.
• Use k-NN classifiers to evaluate predictive accuracy
• Use n-fold cross-validation or leaving-one-out to evaluate predictive accuracy
• Use one- or two-level validation (learning/testing data split)
• Normalize attributes' domains linearly
• Work on any attribute-based data set (presented in a *.nni or *.rde format)
• Be run in parallel with linear speed-up
• Present examples in attribute space in 1, 2 and 3 dimensions graphically

...