This paper addresses the issue of identifying important input features in building an intrusion detection system (IDS). Elimination of the insignificant and useless inputs leads to a simplification of the problem, faster and more accurate systems. The aim is to achieve maximum detection accuracy and to minimize computational complexity. We use algorithms like Ant Colony Optimization (ACO), Genetic Algorithm (GA), Recursive Frequent Elimination (SVM-RFE) and Minimal-Redundancy-Maximal-Relevance (mRMR) for identifying important set of features from the IDS data. We have also used information gain and chi square statistics tool of WEKA (machine learning toolset) to rank the features and identified the important ones. Ant Miner is used for classification and prediction of the data after the feature selection. After identification of the relevant features, the input data is taken as per the reduced feature subset and the optimum parameters for Ant Miner are selected. The experiments were performed using KDDCUP-99 dataset, which is the benchmark dataset for intrusion detection. The prediction/detection accuracy achieved and the time taken for training and testing after use of each hybrid system on the data is finally compared.