Divb30/network-intrusion-detection

Network Intrusion Detection System (End-to-End ML + MLOps Project)

An end-to-end Machine Learning pipeline for detecting network intrusions using real-world MLOps techniques. This project goes beyond model training — it includes a complete training pipeline, modular architecture, data validation with drift detection, MLflow integration via DagsHub, and a FastAPI-based deployment layer.

Key Features

• —Statistical model selection: Bootstrap (n=100) + McNemar's test (p<0.05)
• —Champion-Challenger framework — promotes only on statistically significant gains
• —SHAP explainability — global feature importance + individual waterfall plots
• —KS-Test drift detection across all features
• —MLflow + DagsHub experiment tracking
• —FastAPI for serving model predictions
• —Streamlit monitoring dashboard

Tech Stack

• —ML: Scikit-learn (Random Forest, Gradient Boosting, AdaBoost, LR, DT)
• —Statistical Rigor: SciPy (McNemar's test, KS-test, Bootstrap CI)
• —Experiment Tracking: MLflow + DagsHub
• —Serving: FastAPI + Uvicorn
• —Monitoring: Streamlit + SHAP