Application Of Several Classification Models In Credit Evaluation Of P2P Borrowers

With the advent of the era of big data,the combination of the Internet and traditional industries become more and more closely,and the traditional financial industry is facing the important transition.The research on the credit risk assessment of P2 P borrowers has developed from the traditional expert scoring method to machine learning and data mining.In this paper,three kinds of classification models were analyzed based on the statistical analysis software R language : the application of logic regression,decision tree and ensemble learning model in the credit evaluation of P2 P borrowers.Firstly,the original sample were carried on data preprocessing,from sampling to solve the serious class imbalance problem,and then divide the samples into training samples and test samples to assess the P2 P borrowers whether default in prediction and forecast.In the logistic regression model,the compound selection method was used to filter the variables,and the variables were selected according to the importance of variables in the decision tree and ensemble learning.In the decision tree and ensemble learning,different algorithms are used to build the model,and the loss matrix is introduced into the decision tree C5.0 algorithm.Finally,the effect predicted of the three different models was compared via the confusion matrix and ROC curve.We concluded that decision tree C5.0 algorithm without introducing loss matrix had the lowest error rate,while when introducing loss matrix,C5.0 has the lowest error rate class A,but its high class B error rate;there is no one model could simultaneously had the three error rate in a minimum loss.The analysis of the C5.0 model showed that the most significant impact on the P2 P borrowers was the history of the borrower's credit situation,the motivation of borrowing and repayment ability.