Bayesian Network Learning Based The GA-K2 Algorithm And Its Application In Personal Credit Rating

Since the human society entered into the era of big data and mobile internet, all walks of life have been filled with all kinds of massive data. In order to obtain useful information from large amounts of data, various data mining algorithms arose. Bayesian network is one of data mining algorithms. It is a graphical reasoning method of uncertainty knowledge research, which reflects the interdependence between variables by directed acyclic graph and conditional probability table. Because of its foundation with a strong Bayesian theory and easy to understand graphics mode, Bayesian network have been widely used in many fields. In the personal credit evaluation, how to evaluate personal credit through large amounts of consumer information, and as a basis for classifying consumer are very important for banks and other financial institutions. A good credit scoring model allows the normal operation of the bank’s credit business, and promotes economic growth. So in this paper, we used the German and Australia credit data and Bayesian network classification model to process personal credit assessment.The classic K2 algorithm in all of Bayesian network structure learning algorithms is difficult to determine the order of the input nodes. Think about this, we construct the GA-K2 algorithm based on the global search ability of genetic algorithm, and regard the classification accuracy as the target to perform optimization operation. German and Australia credit data were used in experiments. First, we used a method that Fayyad proposed to discrete data. Then the data samples were divided into 10 parts by the method of K-fold cross-validation, and used the average value as the final result to evaluate the performance of the model. Finally, compared the results based German and Australia credit data with some Bayesian models’ including GA-K2 algorithm, Naive Bayes classifier, Bayesian network classifier based on non-optimized K2 algorithm, TAN classifier; some artificial neural network models’(ANN); support vector machine(SVM) models’ and some other models’ including K- nearest neighbor classifier, Logistic model, decision tree model. The empirical shows that the best accuracy of Bayesian network classifier based GA-K2 algorithm is 82% in German credit data experiments, and the average value reaches 78.5%. In Australia credit data experiments, the best accuracy is 94.2%, and the average value reaches 91.16%. In all of the models, the accuracy of the Bayesian network classifier based GA-K2 is the best, which illustrates the effectiveness of GA-K2 algorithm.The GA-K2 algorithm regarding the accuracy as the target not only solves the problem of input node order, but also optimizes the entire network structure and parameter learning process, also obtains a good result.