DEEP LEARNING-BASED CYBER ATTACK PREDICTION USING DNN-LSTM

Abstract

In this research, new hybrid feature selection framework is proposed, also that can be used to detect Distributed Denial of Service (DDoS) attacks in Software-Defined Networks (SDN). The framework consists of two different approaches: filter-based and wrapper-based. The filter-based feature selection methods use both a local and global subset of features to classify the attack, while the wrapper-based technique uses the Sparrow Search Optimization Algorithm to select performance-based subsets of features from the training data. Both techniques provide a list of common features that will be used for training Machine Learning (ML) and Deep Learning (DL) models. In the performance evaluation of the feature selection phase, the features from the selected subset were used to train the following ML algorithms: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest Classifier (RF), Convolutional Neural Network (CNN) and Ensemble-Based models. The experimental results demonstrate that the hybrid feature selection framework contributes to a reduction in the feature set dimension while achieving high detection accuracy (99.85% accuracy with RF) as compared to the other classifiers tested with 5-fold and 10-fold Cross Validation. The Precision-Recall analysis and ROC analysis performed on the 5-fold and 10-fold cross-validation set support the robustness and generalization capabilities of the proposed hybrid feature selection process for real-time SDN security applications.