Machine Learning-Based Approaches for Detecting and Mitigating Distributed Denial of Service (DDoS) Attacks to Improved Cloud Security

Purpose: The research focuses on detecting and mitigating Distributed Denial of Service (DDoS) attacks in cloud environments. It aims to evaluate the effectiveness of machine learning models, particularly the CNN-LSTM hybrid model and the ID3 decision tree, in ensuring cloud security. Materials and Methods: For this study, the CIC-DDoS2019 dataset was used as the primary source of data. The dataset was divided into training and testing sets using an 80:20 split to ensure robust evaluation. Two models were selected for comparison: the CNN-LSTM hybrid model and the ID3 decision tree. The CNN-LSTM model was designed to combine the strengths of convolutional neural networks for spatial feature extraction with long short-term memory networks for sequence learning, while the ID3 decision tree served as a baseline algorithm to evaluate how a simpler, rule-based approach performs against advanced deep learning architectures. Findings: The experimental results showed that the CNN-LSTM hybrid model significantly outperformed the ID3 decision tree method. Specifically, the CNN-LSTM model achieved a recall of 0.97, precision of 0.98, and an F1-score of 0.98, with an overall accuracy of 98.5% in detecting DDoS attacks. Its superior performance can be attributed to its ability to integrate spatial feature extraction and temporal sequence learning effectively. In contrast, the ID3 decision tree model delivered below-average results when compared to the CNN-LSTM, although it remained a usable solution in certain scenarios due to its simplicity and ease of implementation. Unique Contribution to Theory, Practice and Policy: The CNN-LSTM hybrid model emerges as a highly effective solution for DDoS detection in cloud environments and should be prioritized when developing advanced security frameworks. However, decision tree algorithms such as ID3 still hold relevance, especially in resource-constrained environments where computational efficiency and model simplicity are critical considerations.