Internet Loan Application Scoring Model Based On Feature Extraction Of Restricted Pozmann Machines

As the most popular active area among all the internet banking section,P2P-net-loan owned the thunder speed's development.Till August 2016,the number of online p2p net-loan platform is 2235,and the number of investment to 351.80 million.At the same time,the number of loanees reached 1353.1 thousands and accumulative volume was 2581.509 billion yuan,of which a single month in August,the overall turnover reached 191.03 billion yuan.40 new on-line platform had met the public,but within the same month,the authority had found 57 shut-down platforms 42 platforms with issues,which also illustrated the amount of closed platform was more than the number of the new on-line platforms.While the cumulative number of problem platforms had reached 1978.All these indicated P2P online-loan was founded in a low threshold of the industry and the risk control system is still far from perfect,which will bring enormous risks and losses to the investors and the community.Therefore,investors and companies need better platforms and tools to identify project risks effectively.The model of P2P was first born in the British and American,and its primary development is based on two major conditions,first,Internet technology makes trading efficiency and second,widely applicable and covered of credit data.Compared with traditional industries,P2P platform company has more advantages in terms of data acquisition and lower cost with higher capacity;based on some popular open source big data processing techniques,like Hadoop,Spark and other,can achieve efficient processing of data.On the other hand,the Internet company's financial data is stored in a large warehouse but not a traditional financial institution data acquisition category(such as web form access record)fields,and they are more prone to large quantities of lightweight loans.That's cause by the scarcity of risk control on the traditional model,which do not have enough satisfactory performance.Therefore,we propose to use RBM model that based on characteristics engineering methods.Without expertise relying,it could do better on credit extract information from the database and enhance the accuracy of the model.