Classification trees with soft splits optimized for ranking

Abstrakt

Soft split in a decision tree combines results from both branches when value of the tested predictor is close to the splitting value. The article compares several methods of determining parameters of soft splits as postprocessiong of classification trees when performance is measured by area under ROC curve.

Experimental results use data sets from UCI Machine Learning repository.

Klíčová slova

• supervised learning
• decision trees
• scoring classifier