A fusion method with self-adjusting weights is proposed

which measures the classifier performance by the confusion matrix

and self-adaptively assigns weights to classifiers based on their outputs.Bigger weights are assigned to reliable outputs so that the decision templates are more credible.For a sample which is prone to be misclassified

besides the similarity between it and the decision templates

the information of the training samples around it are included to make a decision.Experiments were done on the KDD’99 intrusion detection dataset and 8 datasets from the database UCI to compare the proposed method with the DT method.The experimental results show the presented method has a better classification performance.