Business failure prediction using distributed ARTMAP neural networks

Predicting corporate failure or bankruptcy is an important task, as it can have significant impact on decision making. This paper proposes a novel approach, based on distributed ARTMAP neural networks. The advantages of these networks over the most popular back-propagation MLPs include fast on-line (one-pass) learning, preserving already acquired knowledge, retention of rare patterns, improvement with experience, and scalability. Conducted experiments using a popular dataset show that in terms of prediction accuracy the distributed ARTMAP is better than other statistical and non-statistical techniques and equally best with an MLP implementation, all applied to the same dataset. An exhaustive search procedure over subsets of financial ratios shows that in contrast to other techniques, the distributed ARTMAP performs its best using two financial ratios only.