Research On Personal Credit Evaluation Model Based On Bayesian Network

In recent years,the proportion that individual unsecured loan accounts for bank loan increases dramatically with the rapid development of society and economy in our country.Individual unsecured loan has become the essential profit point of every commercial bank in China.From the perspective of commercial banks and other financial institutions,the personal credit risk has become the main risk in the banking industry with the individual unsecured loan business continues to expand.We should control the personal credit risk,or it may lead to turbulence in China's financial system in severe cases.Personal credit assessment is an important part in the construction of personal credit system,and it is also the key of personal risk management.Based on the analysis of overseas and domestic research status about the personal credit evaluation,this paper adopted the machine learning method,put forward a quantitative credit evaluation model of personal commercial loan based on Bayesian network theory.This paper took into account the accuracy,efficiency and different application fields,and then established three personal credit evaluation models which respectively based on tree augmented Bayesian network,Markov blanket and feature selection.The paper empirically studied the three models based on the open data "Germany's credit card data set" which is from a universal recognized institute in the machine learning science community.The results showed that the three models are feasible,effective,but different in accuracy and efficiency.The tree augmented Bayesian network model has the highest accuracy and the lowest efficiency,on the contrary,the feature selection model has the lowest accuracy and the highest efficiency.The Markov blanket model is in the middle.At last,the application scenarios of the three models are analyzed and recommended.If the data attribute is obvious,the amount is not much and the demanded accuracy rate is high,you can choose the tree augmented Bayesian network model,and the model is suitable for the traditional small and medium distributed data processing environment.If the original data attribute is numerous,the demanded accuracy rate is not high,you can choose the feature selection Markov blanket model,and this model is suitable for the modern large centralized data processing environment.The Markov blanket model is between the two kinds of model application scenarios.The paper introduced and empirically studied three personal credit evaluation models which respectively based on tree augmented Bayesian network,Markov blanket and feature selection Markov blanket.The three models had broad application prospects.The next objective of the research will be converting the relevant theoretical model into a specific application software prototype.