🔐 Network Intrusion Detection

Classical Machine Learning–Based Intrusion Detection System (IDS)

📌 Project Overview

This project focuses on Network Intrusion Detection (IDS) using structured network traffic data. The objective is to detect malicious or abnormal network activities by building a robust, interpretable, and well-evaluated machine learning pipeline.

The project follows a classical ML–oriented workflow, emphasizing:

• Data integrity and preprocessing correctness
• Careful feature handling
• Fair model comparison
• Clear evaluation and diagnostics

The workflow is designed in a Kaggle-style, ML Engineer–oriented manner, where preprocessing and modeling are separated into dedicated notebooks for clarity and reproducibility.

---

🎯 Objectives

• Prepare clean and model-ready network traffic features
• Design a reproducible preprocessing pipeline
• Train multiple classical ML models suitable for tabular IDS data
• Evaluate models using proper classification metrics
• Compare models to identify the most effective approach for intrusion detection

---

🧠 Problem Formulation

• Task: Intrusion Detection
• Type: Classification (binary or multiclass)
• Input: Structured network traffic features
• Output: Intrusion / attack label
• Evaluation Metrics:
  • Accuracy
  • Precision
  • Recall
  • F1-score
  • Confusion Matrix

---

🧪 Models Used

The following classical machine learning models are implemented and compared:

• XGBoost
• CatBoost
• Random Forest
• Multi-Layer Perceptron (MLP)

Each model is evaluated using:

• Classification Report
• Confusion Matrix
• Metric-based comparison table

Deep learning models are intentionally not used, as classical models are often more effective and efficient for structured IDS data.

---

🗂️ Project Pipeline

The project is organized into clear stages, each implemented in a separate notebook:

Stage	Notebook	Description
1	01_preprocessing.ipynb	Data cleaning, encoding, scaling, and feature preparation
2	02_modeling.ipynb	Model training, evaluation, and comparison

This separation ensures:

• Cleaner notebooks
• Easier debugging
• Reproducible experiments

---

📊 Key Highlights

✅ Structured preprocessing pipeline
✅ Classical ML models tailored for tabular data
✅ Fair model comparison under the same data split
✅ Clear evaluation with confusion matrices & reports
✅ Simple, readable, and reproducible workflow

---

🛠️ Tech Stack

• Python
• NumPy, Pandas
• Scikit-learn
• XGBoost
• CatBoost
• Matplotlib, Seaborn

---

📌 Notes

• Dataset files are not included in this repository
• This project prioritizes engineering clarity over excessive complexity
• The workflow can be easily extended with:
  • Feature selection
  • Hyperparameter tuning
  • Ensemble methods

---

👤 Author

Ardiyanto
Background: Statistics & Machine Learning