Research On Credit Risk Assessment Based On Long Short-term Memory Neural Network Model

With the continuous development of China's economy,the Credit Loan Market also expands at a high rate which lead to the birth of lots of Credit Loan Platforms.Nowadays,when more and more Micro and Small Enterprises and Self-employed Industrialists face money shortage crisis,they need to apply a mortgage or loan from these Credit Loan Platforms in order to get the money they need.As a result,the Credit Risk Assessment become one of the most popular research area,and countless branches of topic has been set up in order to help others to understand the market clearly.Of all the topic,Credit Risk Assessment of the market is the top study area,and the fundamental of any Credit Risk Assessment is the classifying of the credit risk of the borrowers.So how to make the classification is the major study area and how to do it has become the first problem to solve.As Machine Learning Method mature and the application of such methods in Credit Risk Assessment become wider and wider,there has been a change of how to control the risk of each loan.It use to be solved by manual auditing or statistical learning,but nowadays it is solved by computational intelligence.There are four stages of how to achieve the risk control of Credit Loan and to make an accurate classification of the risk through computational intelligence: digitizing the information the borrowers provided and recording the information into different datasets to achieve information digitalization at first place;applying valid feature extraction method on these datasets to pick out the important information which may affect the result of the classification;applying proper classification model to the datasets;and comparing the result with some of the reliable classification models to make a contrast test in order to test the effectiveness and efficiency of the model.In this article,the original controlling method which was used in risk management of Credit Loan is modified through the improvement of a couple of things: the classification model;the statistical processing of digital information;the feature extraction of the dataset and the examination and optimization of the dimensionality of the data.Also there is a proposal of a controlling strategy about the risk management of Credit Loan based on improved Long ShortTerm Memory(LSTM)Neural Networks.In the classification model section,this article started by the studying of the LSTM,surmised the advantage and disadvantage when LSTM Neural Network use Gradient Descent as learning method;then provide a improvement on the learning method use the Conjugate Gradient Method.Also in this article,the experimental data was been given a preliminary audition through the quantitative analysis of the massive amount of information which was provided by actual borrowers;after that the information was summarized and classified into dataset.The importance of how different data dimensions affect the result of classification will then be acquire through the feature extraction of the dataset using Radial Basis Function Neural Network;followed by the amendment of the dataset information plus the optimization and combination of the data dimensions through the analyzation of the correlation of data dimensions and the analyzation of the pseudo-information.At the end of this article is a multiset contrast and evaluation experiment of the risk control strategy of Credit Loan which was described the article.The innovativeness of this article is the use of Conjugate Gradient method in the improvement of the learning method of LSTM,it overcomes the slow local convergence speed which happens during the learning process of the neural network.It also reduces the number of iterations of learning and training,lower the running time of the algorithm and the training error,and increase the accuracy of the classification to some extents.The use of Radial Basis Function during the feature extraction of the datasets and the use of dimension optimization combination of the dataset also help the reduction of redundant information and increase the accuracy of the data information.The multi-set contrast and evaluation experimental helps to verify the affection of risk control strategy in the risk classification area of Credit Loan Market.