|
|
Balancing Node |
Some data mining algorithms, such as decision trees, perform better if classes of records are more evenly balanced (for example: 500 records in class YES and 500 records in class NO, as opposed to 100 in class YES and 900 in class NO). The balance node will duplicate or delete records accordingly to achieve the requested ratio of one class of records to another. The final number of records required can be set to force the program to bias, deleting records from a majority class or duplicating records from a minority class.
The specified percentages should sum to 100; if they do not, WITNESS Miner adjusts them all in a similar way so that they sum to 100%.
Full details of the options available for the balancing node can be found on the balancing options page.
