Multilevel Dimension Reduction for Credit Scoring Modelling and Prediction: Empirical Evidence for Greece. Issue 4 (2nd October 2021)

• Record Type: Journal Article
• Authors: Giannouli, Panagiota; Karagrigoriou, Alex; Kountzakis, Christos E.; Ntotsis, Kimon
• Is Part Of: Communication in statistics. Volume 7:Issue 4(2021)
• Page Start: 545
• Page End: 560
• Publication Date: 2021-10-02
• Subjects: Akaike information criterion -- credit behaviour data -- credit scoring -- elastic net -- multiple dimension reduction -- principal component analysis; Mathematical statistics -- Data processing -- Periodicals; 519.505
• DOI: 10.1080/23737484.2021.1936690
• Languages: English
• ISSNs: 2373-7484

Abstract

Abstract: Several works concerning the modelling and the prediction of credit scoring have been made over time, based on features used in credit scoring, the effectiveness of different classification algorithms and also benchmarking studies classification algorithms for credit scoring. The objective of this work is the proposal of an innovative approach to flexible and accurate credit scoring modelling with the use of not only financial but also credit behavioural characteristics. In addition, we propose a multidimensional reduction algorithm in order to divulge the statistically significant variables that prevail and as an extension to create a reliable prediction model for credit scoring based on the effective combination of principal components analysis and regularization methods. The proposed novel procedure is applied to the Greek System separately for small and large enterprises with the use of a Credit Bureau database with more than 200, 000 cases.