Decision Tree Performance: Impact of Data Sampling Methods

This repository contains an academic project analyzing how different data preprocessing and sampling techniques affect the performance of a Decision Tree classifier. The primary focus is on detecting Network/DDoS attacks using the highly imbalanced CIC-IDS-2018 dataset.

📊 Project Overview

Real-world cybersecurity datasets are often heavily imbalanced, where normal traffic vastly outnumbers attack instances. This project evaluates three sampling approaches to address this issue:

1. Random Sampling (Baseline) - Original imbalanced distribution.
2. Undersampling (RandomUnderSampler) - Reducing the majority class.
3. Oversampling (SMOTE) - Synthetically generating minority class instances.

A detailed Polish report is available in the docs/ folder.

🚀 Key Findings

Based on the experiments, SMOTE oversampling proved to be the most effective method:

• Achieved 99.93% overall accuracy.
• Perfectly classified (100% Precision, Recall, and F1-score) the severely underrepresented minority class (DDOS attack-LOIC-UDP), which originally had only 15 instances in the raw dataset.
• Produced the most balanced and representative logical rules for the Decision Tree.

📁 Repository Structure

├── data/
│   ├── README.md                # Data directory documentation
│   └── download_data.py         # Automated Kaggle dataset downloader
├── docs/
│   ├── README.md                # Documentation overview
│   └── project_report_PL.pdf    # Comprehensive project report (PL)
├── src/
│   ├── README.md                # Source code documentation
│   ├── oversampling_smote.py    # SMOTE implementation
│   ├── random_sampling.py       # Baseline classification
│   └── undersampling.py         # Random Undersampling implementation
├── .gitattributes               # Git attributes configuration
├── .gitignore                   # Git ignore file
├── README.md                    # Main project documentation
└── requirements.txt             # Python dependencies

⚙️ Setup & Installation

1. Install Dependencies

Clone this repository and install the required Python packages:

pip install -r requirements.txt

2. Configure Kaggle API (For Dataset)

This project uses an automated script to download the specific ~300MB CSV file (02-21-2018.csv) from the 7GB CSE-CIC-IDS2018 Kaggle dataset.

1. Create a Kaggle account.
2. Go to your Account Settings -> API -> Create New Token.
3. Place the downloaded kaggle.json file in your system's Kaggle folder (e.g., ~/.kaggle/kaggle.json on Linux/Mac or C:\Users<User>.kaggle\kaggle.json on Windows).

3. Download the Data

Run the automated downloader script from the root directory:

python download_data.py

The script will automatically download the archive, extract the required CSV file into the data/ folder, and clean up the temporary files.

💻 Running the Experiments

Once the data is downloaded, you can run any of the models from the root directory.

To test the SMOTE oversampling method:

python src/oversampling_smote.py

To test the random undersampling method:

python src/undersampling.py

To test the baseline random sampling:

python src/random_sampling.py

👤 Author & Contact

Kacper Lis
Student of Data Engineering and Analysis

• 
• 
• chodzoncy.nieparadox.exe