Nowadays, with the development of internet technologies service in the world, the intruders has been increased rapidly. Therefore, the advent of Intrusion Detection System (IDS) in the security of networks field prevents intruders from having access to the information. IDS plays an important role of detecting different types of attacks. Because network traffic dataset has many features, the process of feature selection and removing irrelevant features increase the performance of the classification algorithms accuracy. This paper provides three various methods which are: Firstly, Information Gain. Secondly, Gain Ratio. Thirdly, Correlation Feature Selection. These techniques are used for selecting and ranking features then select combine the best top ranking features. Only six features were selected out of 41 features. These features are tested on three classifiers (K-Nearest Neighbor, Naïve Bays and Neural Network based Multilayer Perceptron) for classification and detect intrusion. The outcome illustrates that a high level of attacks classification accuracy can be accomplished by combing best different features selection. Moreover, K-Nearest Neighbor gets high accuracy classification for IDS. The proposed model has been applied on KDD data set and ten cross validation process used to assess the classification performance.