| Summary: | 碩士 === 銘傳大學 === 資訊工程學系碩士班 === 102 === With the rapid development of personal loan business, the credit score card is being widely used by banks as a critical technique for measuring the credit risk of a given customer application to accomplish the credit process faster and more accurately. The objective of credit scoring is to predict the default probability of a given customer, convert into a score card score to decide whether the application can be accepted, according to the bank’s risk preference. In recent years clustered-based methods have been proposed to increase the predicting power of credit scoring model, such as using the k-means clustering method to build up a two-stage scoring model. First, a clustering method will be applied to split the raw training data into multiple groups. Second, build models for each of these groups. Last, the scoring process, the new customer should be applied the same clustering built by the first process. And then use the specified scoring model to calculate its credit score. Experiments show an improvement in the classification accuracy. But in real world such improvements are unstable because it can easily cause over-fitting problem. This study would like to improve the clustering and scoring steps in the construction of the credit scorecard to achieve a better prediction performance.
|
|---|
