Machine learning for security
Two models, one lab bench
Classifiers trained on public security datasets and served as live web services. Pick one to try it on your own input.
live
Phishing URLs
PhishGuard
Paste a web address and the model judges whether the domain name looks like phishing.
- DatasetPhiUSIIL
- ModelLogistic regression
- Features17 domain signals
- Accuracy87.9%
Network intrusion
FlowGuard
Upload a CSV of network flows and every flow is scored as an attack or normal traffic.
- DatasetCICIDS2017
- ModelRandom forest
- Features77 flow features
- Macro F10.9997
Both scores come from held-out test data drawn from the same captures as training, so they read higher than these models would manage on traffic from a different network. Educational use only.