Bachelor Project
Skills Gained
Developing Robust Neural Network Models for Network Intrusion Detection: A Comparative Study of Standard and Differentiable Logic Approaches
Deep Learning for Cybersecurity
Neural Networks & Model Comparison
Intrusion Detection Systems (NIDS)
Model Evaluation & Optimization
Applied Machine Learning Research
Neural Networks & Model Comparison
Intrusion Detection Systems (NIDS)
Model Evaluation & Optimization
Applied Machine Learning Research
2025
This bachelor project focused on improving network intrusion detection using neural networks and differentiable logic approaches. We developed and compared a baseline neural network with a differentiable logic–enhanced model trained on the CSE-CIC-IDS2018 dataset containing real DDoS attack traffic. The differentiable logic model achieved stronger overall performance, including higher ROC-AUC and a substantially lower false-positive rate, while also improving robustness and generalization to unseen attack variants. The results demonstrate how integrating logical constraints into neural networks can enhance both accuracy and reliability in real-world cybersecurity systems.
This bachelor project focused on improving network intrusion detection using neural networks and differentiable logic approaches. We developed and compared a baseline neural network with a differentiable logic–enhanced model trained on the CSE-CIC-IDS2018 dataset containing real DDoS attack traffic. The differentiable logic model achieved stronger overall performance, including higher ROC-AUC and a substantially lower false-positive rate, while also improving robustness and generalization to unseen attack variants. The results demonstrate how integrating logical constraints into neural networks can enhance both accuracy and reliability in real-world cybersecurity systems.