A HYBRID METAHEURISTIC METHOD IN TRAINING ARTIFICIAL NEURAL NETWORK FOR BANKRUPTCY PREDICTION

Corporate bankruptcy prediction is an important task in the determination of corporate solvency, that is, whether a company can meet up to its financial obligations or not. It is widely studied as it has a significant effect on employees, customers, management, stockholders, bank lending assessments, and profitability. In recent years, machine learning techniques, particularly Artificial Neural Network (ANN), have widely been studied for bankruptcy prediction since they have proven to be a good predictor, especially in financial applications. A critical process in learning a network is weight training. Although the ANN is mathematically efficient, it has a complex weight training process, especially in computation time when involving a large training data. Many studies improved ANN’s weight training using metaheuristic algorithms such as Evolutionary Algorithms (EA), and Swarm Intelligence (SI) approaches for bankruptcy prediction. In this study, two metaheuristics algorithms, Magnetic Optimization Algorithm (MOA) and Particle Swarm Optimization (PSO), have been enhanced through hybridization to propose a new method MOA-PSO. Hybrid algorithms have been proven to be capable of solving optimization problems faster, with better accuracy. The MOA-PSO was used in training ANN to improve the performance of the ANN in bankruptcy prediction. The performance of the hybrid MOA-PSO was compared with that of four existing algorithms. The proposed hybrid MOA-PSO algorithm exhibits promising results with a faster and more accurate prediction, with 99.7% accuracy.