A Machine Learning Pattern for CNNs with Mixed-Feature Modes

Convolutional Neural Network (CNN) is a Deep Neural Network (DNN) that is well-suited to local feature extraction. DNN learns features useful for a particular learning problem from raw input data, thus reducing reliance on domain experts. However, automatic learning of features via DNN comes at a cost, as it requires relatively high computing resources. An approach to reduce DNN’s resource requirement is to use additional hand-crafted features as input, providing a compromise between automation and resource demand. In this paper, we present a hybrid DNN with mixed-mode features, and demonstrate its application in intrusion detection using UNSWNB15 dataset.