Credit Risk Evaluation System on Big Data using Neurorule

The phrase credit analysis is used to depict any process for evaluating the credit feature of the counterparty. Before offering credit, a bank or other loaner acquires information about the party asking a loan. In the case of a bank offering credit, this might involve the party's annual earnings, existing loans, whether they lease or own a home, etc. Based upon the credit score, the loaning body will decide whether or not to offer credit. The difficulty of credit-risk evaluation is a very questioning and important monetary analysis problem. In recent times, researchers have found that neural networks carry out very well for this intricate and unorganized problem when compared to more established statistical approaches. A major fault associated with the use of neural networks for conclusion making is their lack of clarification capability. While they can achieve a high anticipating accuracy rate, the reasoning behind how they reach their decisions is not easily available. In this paper, the results from analyzing real life credit-risk evaluation data set using neural network rule extraction methods are presented. The dataset chosen is Statlog German credit data set which is readily available for public use at the Statlog repository. Hadoop MapReduce has been used and is a software framework for easily writing applications which handle huge amounts of data in terms of multiterabyte datasets in parallel on large bundles i.e. many thousands of nodes of asset hardware in a dependable, fault-tolerant manner have been used. It has been carried out using Eclipse. This is the reason MapReduce programming is used to work with big data. In addition, Partitioning Around Medoids Clustering algorithm (PAM algorithm) is implemented using map-reduce programming in what way, the medoids are the standards of the neural network to examine credit risk which cluster the customers into the clusters pertaining to the different neurorules. Thus, this paper is aimed to show the derivation of rules from educated neural networks and representing these rules and using it to train large data sets in order to support a reasonable and valuable way for building credit-risk evaluation expert systems.