NEURAL NETWORK APPROACHES FOR DDOS ATTACK EFFECT ASSESSMENT: A SYSTEMATIC REVIEW

Computer Engineering

Distributed Denial of Service (DDoS) attacks represent one of the most critical and persistent threats to modern network infrastructure, targeting essential services and critical systems worldwide. Traditional DDoS attack effect evaluation methods rely heavily on statistical approaches that suffer from limitations including data redundancy, inability to capture complex feature correlations, and dependence on manual parameter tuning. These shortcomings significantly impact the accuracy and reliability of attack assessment, hindering effective defense strategy deployment. This systematic review examines the evolution from traditional evaluation techniques to neural network-based approaches for DDoS attack effect assessment. We comprehensively analyze univariate and multivariate evaluation methods including Index Evaluation Method (IEM), Weighted Sum Method (WSM), Analytic Hierarchy Process (AHP), Grey Relational Analysis (GRA), and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), identifying their inherent limitations in handling modern attack patterns. The review then explores deep learning architectures including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, attention mechanisms, and hybrid models that have emerged as promising solutions. Through comparative analysis of recent studies utilizing datasets such as KDD99, NSL-KDD2009, CIC-IDS2017, CIC-IDS2018, and CIC-DDoS2019, we demonstrate that neural network approaches achieve significantly higher accuracy rates, with state-of-the-art methods reaching up to 99.84% detection accuracy compared to traditional methods averaging below 75%. Feature selection techniques including distance entropy-based Triplet networks, Principal Component Analysis (PCA), and information gain methods are critically evaluated for their role in improving model performance. The review highlights key challenges including slow DDoS attack labeling, computational complexity, model generalization across diverse traffic types, and the need for real-time detection capabilities. We identify future research directions encompassing adversarial robustness, explainable AI for security applications, federated learning for distributed defense, and integration with Software-Defined Networking (SDN) environments. This comprehensive analysis provides researchers and practitioners with insights into selecting appropriate evaluation methodologies and developing next-generation DDoS defense systems

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.