Ensemble GradientBoost for increasing classification accuracy of credit scoring

Abstract

The method for Credit Scoring has been developed to select a better model in predicting credit risk. Data mining methods are superior to the statistical methods of dealing with Credit Scoring issues, especially for nonlinear relationships between variables. By flashing the ensemble method with statistical methods, proven to achieve a higher level of accuracy than the method of data mining. This paper proposes a credit scoring algorithm using Ensemble Logistic Regression by boosting the method using the GradientBoost algorithm. Two datasets for implementing the algorithm, i.e., German and Australian Dataset. The results showed that GradientBoost Ensemble managed to improve the performance of a single classification Logistic Regression and achieve the highest level of accuracy in both datasets. The proposed method produces accuracy of 81% for German datasets and 88.4% for Australian datasets.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint