In this paper, we demonstrate the performance of a data reduction algorithm that is based on the previously itroduced combined bayes test. the combined Bayes test was developed using a noninformative prior on the symbol probabilities. The data reduction algorithm uses a "greedy" approac that relies on the conditional probability of error for the combined Bayes test as a metric for making data reducing decisions. Performance ofj hte algorithm is compared to a neural network at classifying discrete feature vectors which contain six binary valued features. In this comparison it is shown that the Bayesian Data reduction Alorithm is more effective than the neural network at improving overall classification performance.
展开▼
