Rise of new threats such as botnets and DDOS attacks. Machine learning powered firewalls and application
What it does
FIlters packets and marks packet as malicious based on the result of a deep learning neural network.
How we built it
We collated packet capture data from various sources. We then used tensorflow and keras to build deep learning neural networks and models. Using a python tool know as scapy, packets sent to the target port were intercepted and passed through the neural network. The result of the packet classification was then displayed on a web UI.
Challenges we ran into
The initial neural network accuracy was unsatisfactory. We realized that this was due to the overfitting of our data. Through various optimization techniques like regularization dropout and batch normalization, we were able to increase the accuracy of our neural network model to 87%.
Accomplishments that we're proud of
- Achieving 87% accuracy within a day
- Downloading, processing and utilizing several gigabytes of packet data
- Persevering despite the numerous failures of our model
What we learned
- How to design a neural network for categorization
- Machine learning is much harder than we thought
- Machine learning is not just a bunch of statistics
What's next for ML-PAINT
- We can collect and utilize more data
- The structure of the neural network could be modified to better suit the data
- More processors and GPUs could be used to allow the faster processing of a large amount of data