Evolving improved neural network classifiers for bankruptcy prediction by hybridization of nature inspired algorithms

Abstract

Bankruptcy prediction is a hard classification problem, as data are high-dimensional, non-Gaussian, and exceptions are common. Nature inspired algorithms have proven successful in evolving better classifiers due to their fine balance between exploration and exploitation of a search space. This balance can be further refined by hybridization, which may provide a good interplay of exploration (identifying new promising regions in the search space to escape being trapped in local solutions) and exploitation (using the promising regions locally, to search for eventually reaching the global optimum). The aim of this paper is to compare the performance of two search heuristics - Particle Swarm Optimization (PSO) and Gravitational Search Algorithm (GSA) – when using alone, or synergically, as a hybrid method, for evolving Neural Network (NN) classifiers for bankruptcy prediction.