There is a constant rise in the number of hacking and intrusion incidents day by day due to the alarming growth of internet. Intrusion Detection Systems (IDS) monitors network activities to protect the system from cyber attacks. Anomaly detection models identify the deviations from normal behavior and classify them as anomalies. In this paper we propose an intrusion detection model using Linear Discriminant Analysis (LDA), chi square feature selection and modified Naïve Bayesian classification. LDA is one of the extensively used dimensionality reduction technique to remove noisy attributes from the network dataset. As there are many attributes in the network data, chi square feature selection is deployed in an efficient manner to identify the optimal feature set that increases the accuracy of the model. The optimal subset is then used by the modified Naïve Bayesian classifier for identifying the normal traffic and different attacks in the data set. Experimental analysis have been performed on the widely used NSL-KDD datasets. The results indicate that they hybrid model produces better accuracy and lower false alarm rate in comparison to the traditional approaches.