Managing Bank Loans By Using Neural Networks

This study aims at recognizing the role of neural networks in deciding administrative decisions in banks. To achieve the aims, the study developed a suggested model that depends on artificial neural networks as a stabilization tool to support loans management decisions. The Descent Conjugate Gradient algorithm is adopted to build the suggested model through checking loan demands according to the various banking instructions. The results showed that using such techniques in administrative business was a success through evaluating loan demands and deciding the most appropriate ones, with the possibility of refusing or accepting the agent’s demand, and also the possibility of deciding the loans which were demanded more than the other types


INTRODUCTION
Managing loans is a major part of banks and funding companies which offer loans to agents to meet their financial needs. This is because it gains profits for the companies and advantages for agents. Thus, one of the most important things that face banks and agents is how fast getting a loan process goes, with taking into consideration the risks that result from offering loans. Fraud in loans is a fast-spreading issue

Loan concept
The loan language: It is the credit and it is intended for those services provided to customers, according to which individuals, institutions and establishments in the community are provided with the necessary funds, provided that the debtor undertakes to pay that money, its interests, the currencies owed on it and the expenses in one go or in installments on specific date. We conclude that borrowing operations depend on three elements: [6] 1-Trust: In order for the trust factor to be achieved, the customer must provide the bank with guarantees whose financial value exceeds the value of the loan. 1. Bank loans are considered the main resource that the bank relies on to obtain its revenues, as it represents the bulk of its uses in addition to the interests and commissions received by the bank, which represent another source of its revenues.
2. Loans enable banks to contribute to and advance economic activity, as they work to create employment opportunities, increase purchasing power, and improve the standard of living of the community.
3. Achieve economic development with regard to foreign loans in order to cover the need for foreign currencies that are used in the import process.
4. Granting loans means, of course, giving confidence to customers.

4-Advantages of Neural Networks in Loan Management
[3] [9] 1. Lower Operational Cost: Lending companies are able to reduce their operational costs more effectively by handling more applications within the same timeframe. Inevitable, Neural Networks is helping to boost profitability and competiveness.
2. Avoidance of Delay: Lending processes are enhanced when delay of applications are greatly reduced Neural Networks helps processors to take quick actions if a processes is not following an established plan.
3. Accuracy: The algorithm mostly is built and tested on several application data to tune its accuracy for delivery.
Neural Networks helps to streamline human verification operations.

5-Conjugate Gradient Descent Neural Network
It is a conjugate gradient algorithm which is known as classic optimization algorithm. Its main idea is to blend the conjugation between conjugate gradient in downgrading and  PAM Clustering goes through two phases, the first phase is called the construction phase, in which the starting objects are selected and chosen randomly to be the mean, and the variance matrix is calculated to group the objects to the nearest medoid, the second phase is the switching phase using a simple swap process. For each recursive field object i and a non-field object j are selected to get the best cluster [7]. Figure   (2) show the PAM phases

7-Practical Section
When bank experts review loan demands, they data. This is the substitution process, which is the second phase that PAM cluster goes through.   Where the data entered comes out as it is without change, so exponential functions were used to improve the outputs during the training process, so that if the value is far from the real output, the value is modified through the use of the PAM cluster.

8-The work steps:
a-Network Training: The weights of the networks are initialized, so the network is configured for training. Training cases are used to adjust the weights by reducing the prediction made by the network, in order to track the best set of weights that reduce the mean squared error, the algorithm allocates the error backwards across the inner layer of the network and the network evolves through a number of execution times and uses the error to adjust the weights.

b-Network Testing:
After completing the training. The network was simulated on the test set (that is, cases that the network had not seen before).
The results were very good; The network was able to classify most of the cases in the test set, It turns out that there is not much difference between the actual output and the target output.

c-Feed back:
Considering the outputs of the neural network as inputs to the second stage of the work, which is the PAM cluster, to get more pure results to get the right decision without any external interference

9-Results:
The results were divided into 68% as training data, while the rest was considered as test data. After executing the model in 3000 cycles, the results showed that the bank offers loans (if cash is available) in the first place to real estate-loan demanders and to those with high income who provide sponsor with stable income, preferably those who have a bank account. In the second place come loan demanders for purposes of small projects, especially those who are between 25 -30 years old. As for agricultural loans, most of banks offer loans to farmers without interest or commission only to encourage farmers because this benefits the country's economy. Figure (4) illustrate this.