Anomaly detection using ML and Keras

This repo contains my work for the ML course on 'Dataset of anomalies and malicious acts in a cyber-physical subsystem'

You can find the original dataset under assignment_material together with the reformatted logs at exported_logs and graphs of the provided data under exported_graphs_assingment and normalized plots under exported_graphs.

3 Models have been evaluated, described in ml_utils.py looping over different data perparation and training parameters. Using LSTM and Conv layers.

Data preparation has been done using pandas and numpy and all the model description and training using keras and tensorflow. I have used Cuda version of tensorflow as I have a gaming graphics card, but it shouldn't be complicated to refit the models for using the non-cuda keras layers. This code is available for evaluation but I won't maintain it. Use it as inspiration if you like.

You can find the results at test_results.