We created GWC Cyber Lab to explore how coding and data science can help people investigate cybersecurity threats. As Girls Who Code members, we wanted to build something practical that shows girls can take on important challenges in technology, and that machine learning can support human judgment without replacing it.
Our project analyzes recorded network-flow data and flags connections whose patterns resemble attacks such as DDoS or brute-force password guessing. We built a Streamlit dashboard where users can upload a compatible flow CSV, review flagged connections, and explore sample traffic in Replay. We trained and compared logistic regression, decision tree, and random forest models using the CICIDS2017 dataset. In our validation comparison, the selected decision tree caught 1,026 of 1,029 attacks and produced one false alarm among 3,792 normal connections.
One of our biggest challenges was preparing network-flow data for machine learning. The dataset has many features, and real data can include missing, malformed, or infinite values. We built shared cleaning and inference logic so the same feature preparation is used for model training and uploaded CSVs. We also compared false alarms and missed attacks, because a model’s accuracy alone doesn’t explain the consequences of its mistakes.
Through this project, we learned how data preparation, model selection, and clear communication all matter when building an AI-powered tool. We also learned to be careful about what our results mean: the model was trained on lab data, analyzes recorded flows rather than live traffic, and may perform differently on other networks. A flag is a reason for a person to investigate, not proof of an attack.
We’d like to evaluate the models on more diverse network data and report performance on a separate final test set. We also want to make it easier for users to prepare compatible flow CSVs from their own authorized network captures, improve explanations of why a connection was flagged, and make the app available online.
Our goal is to keep growing GWC Cyber Lab into a friendly place to learn about cybersecurity, machine learning, and how thoughtful code can help protect people.
Built With
- cicids2017
- llm
- matplotlib
- numpy
- pandas
- python
- scikit-learn
- streamlit