How buseyaren/classification-and-detection-ddosattacks works
Project type and purpose
• This is a data-science/machine-learning project for detecting DDoS attacks. It uses classical ML (Decision Trees, SVM), reinforcement learning, and deep learning (RNN, LSTM, BiLSTM) methods, all orchestrated via Jupyter notebooks and a Python script.
File structure
• Notebooks:
• Desicion Trees.ipynb (Decision Tree experiment)
• Detection of DDOS Attack.ipynb (overall workflow/demo)
• Training_RNN_DDOS_Attack.ipynb (RNN/LSTM training)
• Script: training_rnn_ddos_attack.py (standalone version of the RNN training code)
• Assets: images/ (plots, model result visuals)
• README.md (project overview, results table)
README for additional insights
• Confirms three comparative approaches: classical ML, reinforcement learning, deep learning
• Shows performance metrics and sample outputs
Main components & interactions (to represent
a. Data Ingestion & Storage
• Raw network packet data (could be CSV/PCAP)
• Stored in a local folder or data lake
b. Preprocessing Module
• Feature extraction (packet length, timestamps, protocol flags)
• Implemented in notebooks (pandas, NumPy)
c. Model Training Engines (three parallel flows)
1. Classical ML Engine
– Uses scikit-learn to train DecisionTreeClassifier and SVM
2. Reinforcement Learning Engine
– Custom RL environment, agent, reward function implemented in Python
3. Deep Learning Engine
– TensorFlow/Keras or PyTorch code in notebooks/training_rnn_ddos_attack.py for RNN, LSTM, BiLSTM
d. Evaluation & Comparison
• Compute accuracy, precision, recall
• Generate plots (matplotlib/seaborn) saved to images/
e. Visualization & Reporting
• Inline charts in notebooks
• Exported images (predicted_attacks.jpg, bwd_packet_length.jpg)
• README.md consolidates key results
Architectural patterns & design principles
• Data‐centric pipeline with clear separation of preprocessing, training, and evaluation
• Modular notebooks/scripts each dedicated to one approach
• Reusable preprocessing code across all engines
• Batch processing (no real‐time streaming)