Most of the intrusion detection systems analyze all network traffic features to identify intrusions with different classification techniques. Any intrusion detection model developed has to provide maximum accuracy with minimal false alarms. Identifying the optimal feature subset for classification is an important task for improved classification. In this paper, consistency based feature selection is used to identify the optimal feature subset. The usage of training data improves the ability to differentiate the classes with similar behavior hence supervised classifiers are deployed to produce more reliable and accurate results. In this paper, we build a hybrid model for intrusion detection integrating consistency based feature selection and ensemble of classifiers such as SVM and LPBoost. The objective of this paper is to determine the relevant features for the construction of model and classify the different network traffic classes using ensemble layer of classifiers. Our experimental analysis using NSL-KDD data set indicate that the proposed ensemble yields high accuracy though there is a huge class imbalance problem in network traffic.