This paper proposes two-level strategy for building the anomaly detection classifier, namely, macro level and micro level classification. The former intend to classify network data on a broader perspective to predict whether it is normal or a potential attack. The later classifies individual anomalies within each category of known attacks. The paper also investigates various feature selection techniques for choosing relevant features and study its effect on the performance of the anomaly detection classifiers. Experiments suggest that employing feature selection along with the proposed approach give anomaly detection rate of up to 99%.