Evolutionary Algorithms for Selecting the Architecture of a MLP Neural Network: A Credit Scoring Case

Neural Networks are powerful tools for classification and Regression, but it is difficult and time consuming to determine the best architecture for a given problem. In this paper two evolutionary algorithms, Genetic Algorithms (GA) and Binary Particle Swarm Optimization (BPS), are used to optimize the architecture of a Multi-Layer Perceptron Neural Network (MLP), in order to improve the predictive power of the credit risk scorecards. Results show that both methods outperform the Logistic Regression and a default neural network in terms of predictability, but the GA are more time consuming than the BPS. The predictive power of both methods is similar to the Global Optimum but it is found in a reasonable time.