The existence of intrusion detection systems is important, because despite all security mechanisms, a system has many vulnerable points which increase the possibility of attack from these system. Therefore, in this research, Intrusion Detection in Computer Networks is considered with a hybrid approach. The proposed method is a three-step approach that attempts to improve the attacks detection and reduce the false alarms. First, the multi-class problem has been converted into several two-class problems and then, appropriate properties of each class is extracted based on a various approaches of information gain and Fisher algorithm. In the last step, we will have one output for each classifier and four outputs for each class. Since the number of dataset classes are examined by five ordinary classes, denial of service attack, port scanning attack, remote local access attack and user attack to the root, and 20 output results of different classifiers are created for each sample. The used data set in this research is KDD-CUP99. The most important used evaluation criteria include precision, readout, false alarm rate, F-criterion, and error rate. In the proposed algorithm, decision tree, naive Bayes, K- nearest neighbor and the neural network have been used as an initial classifier and an incremental algorithm based on the decision tree has been used as the final classifier. The proposed method in all classes could be able to perform better function than the previous method.
