Ryu python SDN controller with Neural Network integration for predicting anomaly in a computer network traffic.
Description
This is a part of SDN project at AGH University of Science and Technology.
The purpose of implementing Deep Neural Network in SDN controller is to predict future bandwidth usage in the network and estimate confidence intervals for decision making algorithm (is there anomaly in the network?).
Neural network model (predict 60s traffic based on 600s sample) LSTM_SDN_seconds_bayesian_60s
Neural Network model (predict 30min traffic based on 5h sample) LSTM_SDN_dropout_wd_30min_sigma
The source for the model is taken from: MachineLearningMastery
Related work (confidence intervals in Neural Networks: uncertainty-sense
Related work (confidence intervals in Neural Networks: Dropout as a Bayesian Approximation
Integration with SDN controller (change measure interval) Ryu integration (anomaly detection)
Related document about time-series anomaly detection Anomaly detection (source)
Environment
Project consists of:
- Open source big data platform PNDA as the main point for network data collection and analysis.
- Ryu python SDN controller
- Mininet open source computer network simulation framework
- iperf3 network traffic generator