Thesis (BS Applied Mathematics) -- University of the Philippines Mindanao, 2008

The loan problem is simply a classification problem where one classifies a loan applicant as a potential good client or as potential bad client, based on the applicant's given characteristics. Common methods used in the classification problem are statistical methods such discriminant analysis, linear regression and logistic regression. Mounting losses due to inappropriate credit choices has lead the use of machine learning methods, such as ANNs, to improve the accuracy of the classification model. The most popular ANN used in the industry is the Back-Propagation Neural Network (BPNN). However, it has been found that despite its popularity the BPNN is not the best ANN. Thus, this study explored the possibility of an Artificial Neural Network (ANN) trained by Differential Evolution (DE) as a loan classifier. This compared the ANN trained by DE's performance in terms of accuracy and training time against the BPNN and a linear classification function trained by the Gradient Descent Search (GDS), and DE. It was found that the best algorithm when the training set has an equal number of positive and negative samples is the ANN trained by DE. Otherwise, the best algorithm was the BPNN. It was also found that the best algorithm that has the shortest training time is the linear classification function trained by GDS because the difference in their speed is hardly noticeable.