A neural approach for SME's credit risk analysis in Turkey
6th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2009, Leipzig, Almanya, 23 - 25 Temmuz 2009, cilt.5632 LNAI, ss.749-759, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 5632 LNAI
- Doi Numarası: 10.1007/978-3-642-03070-3_56
- Basıldığı Şehir: Leipzig
- Basıldığı Ülke: Almanya
- Sayfa Sayıları: ss.749-759
- Anahtar Kelimeler: Credit risk analysis (CRA), K-nearest neighbor (k-NN), Multilayer perceptron (MLP), Neural rule extraction (NRE), Small and Medium Enterprises (SMEs), Support vector machines (SVM)
- Boğaziçi Üniversitesi Adresli: Evet
Özet
This study presents a neural approach which cascades a neural classifier which is multilayer perceptron (MLP) and a neural rule extractor (NRE) for real-life Small and Medium Enterprises (SMEs) in Turkey. In feature selection stage, decision tree (DT), recursive feature extraction (RFE), factor analysis (FA), principal component analysis (PCA) methods are implemented. In this stage, the RFE approach gave the best result in terms of classification accuracy and minimal input dimension. Then, in classification stage, a MLP that is used for preprocessing is followed by a NRE. The MLP makes a decision for customers as being "good" or "bad" and the NRE reveals the rules how the classifier reached at the final decision. In the experiments, Turkish SME database has 512 samples. The proposed approach compared with k-NN and SVM classifiers. It was observed that the MLP-NRE was slightly better than SVM and local k-NN. © 2009 Springer Berlin Heidelberg.