A Novel Cyber Security Model Using Deep Transfer Learning

Abstract

Preventing attackers from interrupting or totally stopping critical services in cloud systems is a vital and challenging task. Today, machine learning-based algorithms and models are widely used, especially for the intelligent detection of zero-day attacks. Recently, deep learning methods that provide automatic feature extraction are designed to detect attacks automatically. In this study, we constructed a new deep learning model based on transfer learning for detecting and protecting cloud systems from malicious attacks. The developed deep transfer learning-based IDS converts network traffic into 2D preprocessed feature maps.Then the feature maps are processed with the transferred and fine-tuned convolutional layers of the deep learning model before the dense layer for detection and classification of traffic data. The results computed using the NSL-KDD test dataset reveal that the developed models achieve 89.74% multiclass and 92.58% binary classification accuracy. We performed another evaluation using only 20% of the training dataset as test data, and 80% for training. In this case, the model achieved 99.83% and 99.85% multiclass and binary classification accuracy, respectively.