Evolutionary Methods for Bankruptcy Prediction: A Study on Indian Firms

Abstract

Corporate bankruptcy prediction has seen impressive research over the last couple of decades, mainly because of its immense importance in credit-risk management. Researchers from academia and industry have nearly exhausted the machine learning space to find the most accurate model for bankruptcy prediction. All of these researches have achieved great yet heterogeneous results. We attempt to extend this research and thereby suggest a homogeneous approach for the bankruptcy prediction problem using evolutionary methods—genetic algorithm (GA) and particle swarm optimization (PSO). We use GA for feature selection and PSO for optimizing the weights and biases in a neural network architecture. The feed-forward neural network is then used for prediction. The novelty of our approach lies in its adaptability and generality. The approach can give promising results for different types of data—from different countries or industrial sectors.