Enhanced SVM model using PCA-autoencoder for DDoS-DNS attack detection in E-commerce networks

Abstract

E-commerce platforms are increasingly targeted by cyber-attacks, resulting in substantial financial losses and damage to their reputation. Many traditional security methods are inadequate at detecting these sophisticated attacks, highlighting the need for smarter solutions. This project addresses this issue by developing a machine learning system specifically designed for detecting intrusions in e-commerce environments. Six model combinations were evaluated, utilizing data simplification techniques such as Principal Component Analysis (PCA) and Autoencoders, alongside classification tools like Support Vector Machine (SVM), XGBoost, and AdaBoost. The models were assessed using the CIC-IDS2017 dataset, which simulates real network traffic scenarios. Their performance was measured based on accuracy, precision, recall, and F1-score. Among the tested models, the Autoencoder-XGBoost combination demonstrated the highest accuracy and the most effective detection capabilities. This suggests that employing deep learning techniques for feature selection, combined with robust ensemble methods, enhances intrusion detection performance. In conclusion, this project demonstrates that machine learning can significantly improve the security of e-commerce platforms when integrated with data simplification and ensemble learning strategies. The developed system offers a more accurate and efficient approach to identifying cyber threats, laying the foundation for future research into more advanced and adaptable cybersecurity solutions.