TL-CNN-IDS: transfer learning-based intrusion detection system using convolutional neural network

Abstract

To address the problems of insufficient training samples and unbalanced sample classes for intrusion detection in real network environments, this paper proposes an intrusion detection system TL-CNN-IDS based on transfer learning and ensemble learning. First, preprocessing using IG-FCBF feature engineering methods followed by conversion of the obtained dataset into an image form suitable for CNN model input. Secondly, three CNN models of VGG16, Inception, and Xception are selected as the basic learning model, and the hyperparameter optimization method of the Tree-Structured Parzen Estimator algorithm is adopted to search the best model on the target dataset. Finally, the optimized CNN model is integrated using the ensemble learning method of confidence averaging. Experiments were conducted on the CICIDS2017 dataset with accuracy, precision, recall, and F1-score exceeding 99.85% and validation of model effectiveness on the NSL-KDD dataset. The experimental results show that the proposed TL-CNN-IDS can achieve network intrusion detection and outperform other intrusion detection methods.