The credit scoring model is a model for delineating personal financial authority, which has emerged in recent years to ensure the financial security of banks and other financial departments. This model refers to obtaining different grades of credit scores according to the customer's credit history data and using a certain credit scoring model, and determining the amount authority that the customer can hold according to the customer's credit score, thus ensuring the security of repayment and other businesses. With the increasingly prominent role of loans and credit cards in modern society and companies, the development prospect of credit scoring model is immeasurable.
2. What are the categories?
(1) Discriminant analysis model
Discriminant analysis is a statistical analysis method to distinguish the category of the research object. For discriminant analysis, it is necessary to know the classification of the observed object and some variable values indicating the characteristics of the observed object. Discriminant analysis is to screen out variables that can provide more information and establish discriminant function, so that the derived discriminant function can minimize the wrong judgment rate when classifying observation samples. The theoretical basis of this method is that the sample consists of two sub-samples with significant differences in distribution, and they have the same properties. It originated from the linear discriminant function introduced by Fisher in 1936. The purpose of this function is to find a combination of variables and distinguish two groups with some common characteristics.
The advantage of discriminant analysis method is suitable for binary or multivariate target variables, and it can judge and distinguish which group an individual should belong to. It also has its own inevitable shortcomings: the model assumes that the distribution of independent variables is normal, but the data in practice are often not completely normal, which leads to the unreliability of statistical results.
(2) Decision tree method
Decision tree model is a statistical technique that divides the population continuously to predict the results of a certain target variable. The input of decision tree construction is a group of examples with category labels, and the result of construction is a binary or multi-branch tree. The method of constructing decision tree is top-down recursive construction. In practice, personal credit is selected as the target attribute and other attributes as independent variables for personal credit analysis. All customers are divided into two categories, that is, good customers and bad customers, and the credit status of customers is converted into whether they are good customers or not (the value is 1 or and then a complete decision tree is generated by using the data set. A rule base can be established in the generated decision tree. A rule base contains a set of rules, and each rule corresponds to a different path of the decision tree, which represents a link of the conditions it passes through the nodes. By creating a decision tree for the best classification and discrimination of the original xiangben, the recursive segmentation method is adopted to minimize the expected misjudgment loss.
Advantages of decision tree model: The shallow decision tree is very intuitive and easy to explain; There is no need to make any assumptions about the structure and distribution of data: it can be easily transformed into business rules. Its shortcomings are as follows: the deep decision tree is difficult to explain visually; Decision tree has a large demand for sample size; Decision trees tend to be too fine-tuned to sample data and lose stability and shock resistance.
(3) Regression analysis
Regression analysis is the most widely used credit scoring model so far, among which the famous logistic regression is the representative. In addition, linear regression analysis, probit regression and other methods also belong to this category. Orgler, who was the first to use regression analysis, made a scorecard similar to a credit card by using linear regression model. His research shows that consumer behavior characteristics can predict the possibility of future default more than the application materials. As in the mathematical programming method, it is assumed that a number of indicators have been extracted from the sample variables by a certain method as the idea of feature vector regression analysis. The idea of regression analysis is to fit these indicators into an explained variable that can predict the applicant's default rate, which is naturally the default rate. The most widely used model in P regression analysis belongs to the linear regression model, which simulates a straight line of the quantitative relationship shown in a large number of data points, and the goal of back-to-day analysis is to minimize the error between the target variable value and the actual target variable value. Therefore, the simple linear regression model is the earliest model that applies regression method to credit scoring research. At present, the credit scoring system based on logistic regression is the most widely used.