DDoS attack detection system using Artificial Neural Network (ANN) / Muhammad Syamil Yusri

• Main Author: Yusri, Muhammad Syamil
• Format: Thesis
• Language: English
• Published: 2024
• Subjects: Neural networks (Computer science)
• Online Access: https://ir.uitm.edu.my/id/eprint/95729/1/95729.pdf; https://ir.uitm.edu.my/id/eprint/95729/

DDoS attacks represent a substantial danger to network availability and security. Because of their sophistication, traditional DDoS detection technologies sometimes struggle to effectively identify and neutralise such attacks. In this project, offer a novel method for detecting DDoS attacks using Artificial Neural Networks (ANNs). To detect aberrant network traffic associated with DDoS attacks, the system uses the capability of ANNs, which are capable of learning complicated patterns and generating accurate predictions. The suggested system is divided into two stages: training and detection. A systematic strategy is used in the study framework, which includes problem identification, data gathering, preprocessing, ANN implementation, and performance assessment. The project's goal is to improve network security by identifying and categorising DDoS attacks properly. The implementation of ANNs using the Sequential model was tested using different training/testing splits (80/20, 70/30, 60/40) and epochs. The 70/30 split regularly outperformed others, with an accuracy rate of 92.29%. Detailed parameter tweaking was carried out for the 80/20, 70/30, and 60/40 splits, finding that a three-hidden-layer architecture with 256, 128, and 8 neurons produced the highest accuracy of 92.54%. Multiple ANN models were evaluated, and the best-performing model achieved 92.54% accuracy, 93.01% precision, 91.05% recall, and a 92.10% F1-score. However, the research recognises several limits, such as the time-consuming nature of ANN model training, possible scaling concerns due to hardware restrictions, and fluctuation in dataset appropriateness. To overcome these constraints, the project propose future recommendations like as using advanced techniques like parallel processing, automated methods for dynamic dataset selection, and hybrid approaches that combine ANN with other methods to increase accuracy.