Neural network classification and prior class probabilities

A commonly encountered problem in MLP (multi-layer perceptron) classificati on problems is related to the prior probabilities of the individual classes - if the number of training examples that correspond to each class varies significantly between the classes, then it may be harder for the network to learn the rarer classes in some cases. Such practical experience does not match theoretical results which show that MLPs approximate Bayesian a poste riori probabilities (independent of the prior class probabilities). Our inv estigation of the problem shows that the difference between the theoretical and practical results lies with the assumptions made in the theory (accura te estimation of Bayesian a posteriori probabilities requires the network t o be large enough, training to converge to a global minimum,. infinite trai ning data, and the a priori class probabilities of the test set to be corre ctly represented in the training set). Specifically, the problem can often be traced to the fact that efficient MLP training mechanisms lead to sub-op timal solutions for most practical problems. In this chapter, we demonstrat e the problem, discuss possible methods for alleviating it, and introduce n ew heuristics which are shown to perform well on a sample ECG classificatio n problem. The heuristics may also be used as a simple means of adjusting f or unequal misclassification costs.