DDOS Attack Classification With Hyperparameter Tuning And Enhanced Prediction Technique

Abstract

Cyber-Physical Production Systems (CPPS) are integral components of Industry 4.0, enabling seamless integration between computational and physical processes. However, the interconnected nature of CPPS makes them vulnerable to cybersecurity threats, such as Distributed Denial-of-Service (DDoS) attacks. The technologies utilized are Machine Learning Algorithms, Rule-Based Systems, Networking Tools and Protocols, Programming Languages and Platforms, Scikit-learn, TensorFlow, or Py-Torch for model development. The current or existing model has only binary classification indicating malicious and normal attacks. And various strategies, including machine learning (ML) techniques, have been employed for DDoS attack detection. Notable approaches include deep learning models, ensemble methods, and feature selection techniques. But this existing model does not store the result logs for future reference. While these methods achieve reasonable accuracy, they often struggle with issues like class imbalance, scalability, and real-time adaptability. In the other hand the model that is being proposed works to enhance DDoS attack multi classification through suitable trained and tested datasets using LGBM algorithm. Here it would classify the normal and malicious attacks into various types of attacks such as DDOS attack, HTTP attack, traffic detection etc. A Flask-based web application for data upload and result display. Designing a database to store results and logs in file format for future use and enhancement. It operates with an average time complexity of O(log n) for both training and inference, making them highly efficient for real-time DDoS detection. The system demonstrates high classification performance, with precision and recall exceeding 99%, while maintaining a computationally efficient structure suitable for real-time implementation.