Analytical Techniques in the Assessment of Credit Risk by Michalis Doumpos & Christos Lemonakis & Dimitrios Niklis & Constantin Zopounidis

Author:Michalis Doumpos & Christos Lemonakis & Dimitrios Niklis & Constantin Zopounidis
Language: eng
Format: epub
ISBN: 9783319994116
Publisher: Springer International Publishing

The development of a classification tree is performed through an iterative process, starting from the root node. Every stage (node) of this process consists of three individual steps:1.Evaluation of the discriminating power of the risk attributes using the training cases falling at the node under consideration.

2.Selection of the most discriminating attribute.

3.Specification of the splitting rule for the selected attribute, which provides the best classification.

This procedure is repeated until a termination criterion is met. At a second phase, a pruning process is often performed, which cuts down parts of the tree, which are too specialized and may lead to poor generalizing performance. This leads to a simpler and more general tree, which is easier to understand and implement.

The above process for tree construction can be implemented with different algorithms. Some typical and well-known examples of such algorithms are C4.5 (and its improved version C5.0) as well as CART (classification and regression trees).

Classification trees and decision rules offer a very convenient and nature modeling form for credit risk assessment, and they are computationally applicable to large data sets. Moreover, they can handle qualitative and even missing data in a straightforward manner, without requiring complex data transformations. However, they are often prone to overfitting and their probabilistic outputs do not have a continuous form, as each leaf node is usually associated with only one estimate.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.