Modeling based on induction of a large set of random rules enables
detection of some relevant properties of the training set. They are:
Attribute importance means identification of attributes that are most useful for the classification of examples. Presented is an ordered list of attribute names. Each name is followed with a number in the range 1-100 denoting the relative importance of the attribute. The numbers correspond to the weighted and normalized total number of induced rules in which the attribute has been used. If attribute file is prepared and apploaded then attributes will be referenced by their real names.
Outliers are training examples that are substantially different from other examples in the same class. In the process of estimation of the classification accuracy on unseen examples they are identified as examples that received more votes for some incorrect class than for the correct class. The number denoted with each example is the difference between these two values and it is used to order the outliers.
Prototypes are training examples that are very representative for the corresponding class. The report includes one prototype for each class. They are identified as examples that received a large number of correct predictions. The denoted numbers present the difference between the number of votes for the correct class and the number of votes for the most voted incorrect class.
The extended report is obtained by setting the checkbox in the data upload form.
© 2013 LIS - Rudjer Boskovic Institute
Last modified: June 29 2015 12:46:33.