Algorithms
Gini-Index:
-
The impurity measure
is gini-index, just binary splits are performed.
-
For categorical attributes:
if (number of categories <= 10), the optimal binary split is performed.
-
For categorical attributes:
if (number of categories > 10), the heuristic is: order the categories
according to their gini-index, find the best binary split for this ordering.
-
Best experimental results
for the parameters: Purity 0.99/ Number of records 5.
Gini/SLIQext:
-
Like Gini-Index except:
For categorical attributes: if (number of categories > 10), the heuristic
is SLIQext -> Coppersmith C., Hong S. J., Hosking J.R.M.: 'Partitioning
Nominal Attributes in Decision Trees', Data Mining and Knowledge Discovery
Journal, Vol. 3, No. 2, 1999.
Look-4-2:
-
looks for one or two
split points, not published yet.