A Study On Credit Risk Assessment Method Based On Decision Tree

Credit risk is an important research subject in financial field. Strengthen commercial bank credit risks management is not only the objective request of consummating socialist market economy, but also the reality request of trailing connection with international and dealing with the new Basel capital agreement. In recent years, along with the credit economy scale's unceasing expanding, credit derivative tool's massive using, particularly the outbreak of the global financial crisis in last year, credit risk once again become the focus of attention of the international academic community.This article uses method which combines standard research with empirical resarch to conducte in-depth study on credit risk assessment method which is the most important part in credit risks management. proposed in the decision tree with data mining technology to build a credit risk assessment model for the basic idea. It analyzes the characteristic of different credit risks assessment methods, points out the existing evaluation model's shortcomings, puts forward fundamental thought of using decision tree in data mining technology to establish credit risk assessment model.The main research results summary is:Firstly, aim at the continuous characteristic of financial data in credit risk research, the dissertation first prove theorem about relationship between cut points of Normalized Gain-based appraise criteria and boundary points'attribute values, and taking this as the foundation a discretization algorithm based on boundary points'attribute values merging and inconsistency checking is proposed. This method break the idea of traditional continual attribute discretization's traversal heavy search, obviously enhanced the separate effect on the bases of efficiency.Secondly, the paper analyzes the limitation of traditional decision tree method existing in credit risks appraisal, proposed a combined optimization and a multiple analysis decision tree algorithms. The former algorithm makes improvements from three aspects: discretization, reducing dimension, attribute selection, which effectively solves the conflict between efficiency and prediction precision. The latter using the decomposition forecast's multi-sorter thought that solves the noise enlargement problem which caused by the multi-categories data superimposition. The effectiveness and efficiency of classification as compared with ordinary methods has greatly improved.Thirdly, constructing the credit risk evaluation models separately on the basis of combined optimization and multiple analysis decision tree algorithms,and providing the practical algorithm procedures by taking advantages of two classification samp1es and multip1e classification samples of the listed companies as well as the German individual credit data. Through comparative analysis with other model's appraisal results, confirmed model proposed by this paper has the remarkable superiority in credit risk assessment.Fourthly, based on the above-mentioned theoretical results, giving a process design to integrated credit risk assessment system, conducts a deep research on the structure, function and interface design of each modules, and gives the detailed credit risk intelligence auxiliary analysis system design plan.