A Comparison of Credit Rating Classification Models Based on Spark- Evidence from Lending-club

Credit evaluation becomes more and more important in the financial field. Based on the data set of Lending Club-an american P2P platform, we apply four classification algorithms: logistic regression, naive bayes, decision tree and support vector machine, and compare their classfication accuracy, AUC(Area Under Curve)and PR(Precision-recall Rate). By changing the number of nodes of Spark cluster, we compare the run time of these algorithms and analyze the variance of the run time of each algorithm with the increase of the number of nodes. Experimental results show: (1) among these algorithms, the classification accuracy of naive bayes and decision tree is over 40% higher than that of logistic regression and support vector machine when the number of nodes is 3;(2) the AUC value of naive bayes and decision tree is over 45% higher than the other two algorithms; (3)as for the PR value, the performance of naive bayes and decision tree is better, the gap is over 40%; (4)In pseudo-distribution state, the run time of naive bayes is the shortest, but the run time of decision tree decreases obviously from 109.0s to 9.8s when the number of nodes increases from 1 to 7. The results provide a reference for credit assessment in the era of big data, and give suggestions from the perspective of classification algorithm and cluster.