Credit risk analysis with machine learning techniques in peer-to-peer lending market

Sammanfattning: After the sub-prime mortgage crisis of 2007 and global crisis of 2008, credit risk analysis has becomes more important than ever before. This paper conducts the credit risk analysis and compares classification performances among different algorithms (logistic regression, support vector machine, decision tree, multilayer perception, probabilistic neural network, Deep Learning) by using a large peer-to-peer lending dataset composed of a million observations. The findings show that Support vector machine (SVM) provides the most accurate performance, followed by decision tree, logistic regression, multilayer perceptron neural network, probabilistic neural network and deep learning. The main contributions of this paper is the reapplication of machine learning techniques to an alternate dataset composed of significantly larger number of observations with deviating pattern from traditional bank loans. The findings from SVM and Decision tree are consistent with the previous literature. The results from logistic regression and MLP indicate that they are identical based on p2p dataset, which makes a contribution to the debate whether MLP out performs logistic regression. For PNN it is difficult to say if it properly accounts for the data imbalance due to the low performance of the model compared to the others. Deep learning performance is in contrast to previous work as it is the worst performing model comparing with other investigated techniques. This is potentially due to the simple approach to deep learning that this paper adopted and opens up the topic for future research.