Study on the Model of Credit Early Warning Based on SOM-PNN
Yan Ping Peng · Mini-micro Systems · 2005
The risk prediction model is very critical and regarded as the core part of the risk early warning system. Compared with other traditional prediction models, the neural network model has the advantages of self-studying, self-organizing as well as self-adapting. This paper presents a SOM-PNN based risk prediction model, which combines the Self-Organizing Map Neural Network and the Probabilistic Neural Network together. Futhermore, the improved iteration ways in constructing and training the model are also presented, which includes the SOM boundary effect processing and the rare samples handling. The established model is trained with financial ratios for a specific credit risk early warning experiment. The preliminary experimental result demonstrates that the SOM-PNN model perfoms better than some traditional ones in the rates of prediction accuracy and efficiency.