Attribute selection based on regularization networks-genetic algorithm and its application in chemical pattern recognition

The Bayes regularization method is employed to get a well generalized neural networks, the redundant and irrelevant attributes can be deleted from the prime attributes set by printing the input node of the networks. In order to reduce the complexity of searching optimal subset, by the classification accuracy and fitting error of networks used as the first and second fitness value separately, we present a heuristic genetic algorithm to prune the networks by the sensitivity analysis, and the minimum with optimal attributes subset which represents the characteristic of classification can be selected from the patterns of high dimensionality. Finally, the problem of attribute selection and patterns classification of spearmint essence are employed to verify the validity of this method, the result shows that the method is superior to other methods obviously, and the method is also useful in chemical data mining.