A Modified 1D-CNN Based Network Intrusion Detection System

Abstract

One of the crucial components of network security is intrusion detection. The well-received detection technology used conventional machine learning techniques to train the intrusion samples. Since the accuracy of intrusion detection is not commendable in traditional ML technologies, in this paper, we investigate and research a deep learning approach for developing a versatile IDS using a one-dimensional Convolutional Neural Network (1DCNN) which is normally used for supervised learning on time-series data. In each experiment, the CNN models are raced up to 10, 20, 30, and 40 epochs. For comparing the performance along with 1D-CNN, SVM and Naïve Bayes techniques are also utilized. 1D-CNN have outperformed compared to other two techniques. This is mainly because of the rationale that CNN has the potential to extract high-level feature representations.