Research Article
A Novel Oversampling Technique for Imbalanced Credit Scoring Datasets
@INPROCEEDINGS{10.1007/978-3-031-35081-8_12, author={Sudhansu Ranjan Lenka and Sukant Kishoro Bisoy and Rojalina Priyadarshini and Jhalak Hota}, title={A Novel Oversampling Technique for Imbalanced Credit Scoring Datasets}, proceedings={Intelligent Systems and Machine Learning. First EAI International Conference, ICISML 2022, Hyderabad, India, December 16-17, 2022, Proceedings, Part II}, proceedings_a={ICISML PART 2}, year={2023}, month={7}, keywords={Credit Scoring Imbalance Class Distribution Noisy Samples Oversampling Classification}, doi={10.1007/978-3-031-35081-8_12} }- Sudhansu Ranjan Lenka
Sukant Kishoro Bisoy
Rojalina Priyadarshini
Jhalak Hota
Year: 2023
A Novel Oversampling Technique for Imbalanced Credit Scoring Datasets
ICISML PART 2
Springer
DOI: 10.1007/978-3-031-35081-8_12
Abstract
The imbalanced class distribution of credit-scoring datasets typically makes the learning algorithms ineffective. In this study, NOSTE is proposed, a novel oversampling technique. It first identifies the informative minority instances by eliminating the noisy samples from the minority subset. Then, weight is assigned to the informative minority instances by considering the density and distance factors. Finally, new minority instances are created by determining the average of two different minority instances to make the dataset balanced. In the experimental study, NOSTE performance is validated by conducting an extensive comparison with four popular oversampling methods using three credit-scoring datasets from the UCI repository. The results confirmed that the proposed method brings significant improvement in the classification in terms of F-measure and AUC (Area under the Curve).
