How sondt1337/ai-based-ics-incident-investigation-system works
Main Components
• AI Module – Represented by the “ai” directory, which includes notebooks, scripts for preprocessing (ai/xgboost_preprocess.py), confusion matrix computation, and various tuning scripts (grid, hand, hyperopt). This component handles data analysis, anomaly detection, and training of the XGBoost model.
• Benchmarking Module – Found in the “benchmark” directory, which holds benchmarking scripts for different models (XGBoost, Logistic Regression, SVM, etc.) to compare performance.
• Data Storage / Collection – The “csv” and “pcaps” directories hold the raw datasets and network packets respectively. CSV files include datasets and summary files, while PCAP files provide recorded network traffic captures.
• Frontend/Visualization – The “frontend” directory contains the Dockerfile, docker-compose, and the Python app responsible for presenting the visualization (network topology, infected devices, etc.). It integrates with the AI analysis module to display detection results.
• Demo & Additional Resources – The “demo” folder provides demo videos that can be referenced to understand the user experience and visualization flows.
Relationships and Data Flow
• AI Training – From preprocessed data, the AI module (specifically the XGBoost training components) processes the data for both training and evaluation. Note any tuning steps as iterative loops for optimizing parameters.
• Benchmarking – Parallel to the training, the benchmark scripts compare AI model performances to validate performance metrics.
• Visualization Component – The results from the AI analysis (malicious detection outputs, network flows) are then fed into the frontend application which visualizes network topology, lists infected devices, and flags malicious actions.
• Deployment – Include containerization via Docker (seen in the “frontend” Dockerfile and docker-compose), which wraps the frontend service for deployment.
Architectural Patterns & Key Design Principles
• Data Flow Pipeline – Reflect a pipeline where raw data (from pcap and csv) is transformed, features are extracted, and then provided as inputs to machine learning algorithms.
• Integration via APIs / Data Sharing – Make clear the connectivity between the AI backend and the frontend visualization layer, likely via REST APIs or direct import.
Relevant Technologies and Frameworks
• XGBoost – As the primary AI model used for multi-class classification.
• Python – The underlying language for the AI modules, benchmarks, and frontend app.
• Docker & Docker Compose – For containerizing the frontend (and possibly other components).
• Possibly additional libraries used in preprocessing, tuning (like Hyperopt), and visualization.
Final Assembly
• Zoom in on the AI component by further breaking it down into XGBoost tuning, confusion matrix evaluation, and performance benchmarking.
• Lastly, include deployment details showing how Docker configurations (frontend/Dockerfile and docker-compose.yml) enable the system to run as a containerized application.