Research problem
Intrusion detection models often report high accuracy on one dataset but perform poorly on new traffic. This project compares classical machine-learning and deep-learning models on public intrusion-detection datasets and studies how well they generalise.
Research stages
- Literature review and research gap
- Choosing datasets and preprocessing (class imbalance, feature selection)
- Baseline models: random forest, gradient boosting
- Deep-learning models: a feed-forward network and an autoencoder
- Cross-dataset experiments to test generalisation
- Statistical comparison of results
What you will understand
- How to frame a research question and a hypothesis
- Designing fair experiments and avoiding data leakage
- Reading and citing research papers
How it is evaluated
- Precision, recall, F1 and false alarm rate per attack class
- Results on a second dataset the model was not trained on
- Significance testing between models
Thesis and paper
- Thesis chapters: introduction, literature review, methodology, experiments, results, conclusion
- Guidance on turning the results into a conference paper draft
Defence questions to prepare
- What is your contribution beyond existing papers?
- How did you avoid data leakage?
- Why did performance drop on the second dataset?