The following node is available in the Open Source KNIME predictive analytics and data mining platform version 2.7.1. Discover over 1000 other nodes, as well as enterprise functionality at http://knime.com.
This is the start node of a backward feature elimination loop. The first iteration of the loop is executed
with all input columns. In the next
n - 1
iterations each of the input columns - except the target column, that must be selected in the dialog - is
left out once. Then the end node will discard the column that influenced the prediction result the least.
Then
n - 2
iterations follow where each of the remaining columns is left out once, and so on. The total number of
iterations is therefore
n * (n + 1) / 2 - 1.
The backward feature elimination loop can handle arbitrary target column, for numerical columns the squared
error is taken as quality measure, for all other columns the error rate (i.e. the fraction of rows for which
target and prediction column have different values).
0 | Any datatable that contains at least three columns |
0 | The input table with some columns filtered out |