Using Statistics to Assess the Performance of Neural Network Classifiers

Neural network (NN) approaches to pattern classification problems both complement and compete with statistical approaches. Each approach has unique strengths that can be exploited in the design and evaluation of classifier systems. In the spirit of emphasizing this complementary nature, four points are made in this article. First, classical (statistical) techniques can be used to evaluate the performance of NN classifiers. Second, classifiers often outperform classical techniques. Third, classifiers may have advantages even when their ultimate pelformance on a training set can be shown to be no better than the performance of a classical classifier. Finally, it is suggested that methods that are routinely used in statistics, but not in approaches, should be adopted for the latter as well.