Research Article
A Pruning Neural Network Model in Credit Classification Analysis
Table 8
Classification accuracy rates comparison between PNN and other algorithms obtained from literatures of the Australian credit dataset.
| Authors (published year) | Algorithms (train-to-test ratios) | Classification accuracy rate (%) |
| Luo et al. (2009) [48] | SVM (10 CV) | 80.43 |
| Peng et al. (2011) [49] | Bayesian network (10 CV) | 85.22 | KNN (10 CV) | 79.42 | RBF network (10 CV) | 83.04 |
| Yu et al. (2011) [1] | C4.5 (10 CV) | 84.3 | LVQ (10 CV) | 82.97 |
| Chang and Yeh (2012) [50] | SVM (10 CV) | 84.7 | C4.5 (10 CV) | 82.5 | Naive Bayes (10 CV) | 84.9 |
| Zhu et al. (2013) [51] | QDA (5 CV) | 80.02 | DT (5 CV) | 83.18 |
| Tsai et al. (2014) [52] | MLP (10 CV) | 82.44 | DT (10 CV) | 84.91 |
| Lessmann et al. (2015) [8] | CART (10 CV) | 66.4 | ELM (10 CV) | 69.8 | LDA (10 CV) | 78.9 | Logistic regression (10 CV) | 80.7 | ADT (10 CV) | 79.8 | Bag (10 CV) | 76.8 | Boost (10 CV) | 81 | Random forest (10 CV) | 85.2 |
| Khashei and Mirahmadi (2015) [7] | QDA (50%-50%) | 80.1 | SVM (50%-50%) | 77.5 |
| Our method (2017) | PNN (50%-50%) | 85.64 | PNN (5 CV) | 85.31 | PNN (10 CV) | 85.19 |
|
|