The average number of remove attributes of all discretization algorithms is summarized in Fig. 11c.
Some discretization algorithms are an implicit coupling with supervised feature selection, especially EMD, ChiMerge, and MChi2 as shown in Fig. 15c, average number of remove attributes.
The results of AUC, number of intervals, number of remove attributes, and running times of 13 discretization algorithms of 20 imbalanced datasets (fivefold) and 4 classifiers.
The results of predictive accuracy, number of intervals, number of remove attributes, and running times of 13 discretization algorithms of 30 standard datasets (tenfold) and 4 classifiers.
The details of the result are included as an Additional file 3. Fig. 15 Discretization results of: a average number of intervals, b average running times (seconds), and c average number of remove attributes for imbalanced datasets.
The details of the result are included as an Additional file 2. Fig. 11 Discretization results of: a average number of intervals, b average running times (seconds), and c average number of remove attributes for standard datasets.
The second change is remove attributes in blocks, rather than one by one.
Furthermore we see that this increase in success rate is not redundant with the tuned approaches, as both of these, TuRF and SURF & TuRF, which iteratively remove attributes with low quality estimates, are much better than the standard Relief and SURF approaches at selecting a small subset which contains the functional attributes.
The second mechanism, adaptive removing attribute (ARA) mechanism, removes some redundant attributes in the process of selecting node.
Given an attribute subset (Bsubseteq A) and a decision attribute (d), the projection of the object set (U) on (B) is denoted by (Pi _{B }(U)) and can be computed by the following two steps: removing attributes in the different set ((A-B)) and merging all the remaining rows which are indiscernible [11].
