For computationally efficient and effective IDS, it is essential to identify important input features. In this paper, a statistical feature construction scheme is proposed in which factor analysis is orthogonally combined with an optimized k-means clustering technique. As a core component for unsupervised anomaly detection, the proposed feature construction scheme is able to exclude the redundancy of features optimally via the consideration of the similarity of feature responses through a clustering analysis based on the feature space reduced in a factor analysis. The performance of the proposed method was evaluated using different data sets reduced by the ranking of the importance of input features. Experimental results show a significant detection rate through a good subset of features deemed to be critical to the improvement of the performance of classifiers.