Using the Boosting Technique to Improve the Predictive Power of a Credit Risk Model
Alejandro Correa, Colombia Darwin Amézquita · 2013
In developing a predictive model, the complexity of the population used to build the model can lead to very weak scorecards when a traditional technique such as logistic regression or an MLP neural network is used. For these cases some nontraditional methodologies like boosting could help improve the predictive power of any learning algorithm. The idea behind this technique is to combine several weak classifiers to produce a much more powerful model. In this paper, boosting methodology is used to enhance the development of a credit risk scorecard in combination with several different techniques, such as logistic regression, MLP neural networks, and others, in order to compare the results of all methodologies and determine in which cases the boosting algorithm increases model performance.