Machine learning enables intrusion detection systems to detect network attacks adaptively and intelligently. Recently deep neural network has been investigated as such a solution owing to its high accuracy but it has limitation in real-Time performance. To enhance the learning time, in this paper, we propose to use feature selection and layer configuration. We use the NSL-KDD data set, which is a refined version of the KDD CUP 99 data set and analyzed the associations between features using WEKA, a data mining tool. Our experimental results confirm that proper feature selection and layer configuration can reduce learning time while maintaining high average accuracy.