🚀 5G Project: SDN Traffic Classification

🌟 Group Number: Gr37EC431

Group Team Members:

• AYUSH KUMAR AGARWAL (202151037)
• PARMAR HET (202151103)
• PATEL SHUBH UMESHKUMAR (202151111)
• POPAT RITIK MANISH (202151114)
• RAJ JIGNESHKUMAR SHAH (202151127)
• JENNY KAMLESHBHAI BHUT (202152314)

---

📖 Overview

This project leverages Software-Defined Networking (SDN) for traffic classification in 5G networks using supervised learning and unsupervised learning techniques. By analyzing traffic patterns and characteristics, it aims to enhance:

• 🛠 Network Efficiency
• 🔒 Security
• ⚡ Resource Allocation

Installation

Follow these steps to set up and run the project locally:

1. Clone the Repository

First, clone the repository to your local machine:

git clone https://github.com/IIITV-5G-and-Edge-Computing-Activity/SDN-Traffic-Classification.git

cd SDN-Traffic-Classification

2. Install dependencies:

pip install -r requirements.txt

🔍 Methodology

🧑‍🏫 A. Supervised Learning with Logistic Regression

1. 🧹 Dataset Preparation:

   • Cleaned and encoded the data.
   • Split into features (independent variables) and labels (dependent variables).
   • Used train_test_split for a 70-30 split between training and testing data.

2. ⚙️ Model Initialization:

   • Chose Logistic Regression for its simplicity and effectiveness in multiclass problems.
   • Configured parameters:
     • Penalty: L2 regularization.
     • Solver: Liblinear.

3. 📊 Training Process:

   • Trained the model using the one-vs-rest approach to handle multiclass classification.

4. 🔄 Cross-Validation:

   • Performed 5-fold cross-validation for robust evaluation.
   • Analyzed metrics across folds to detect overfitting or underfitting.

5. 📈 Evaluation Metrics:

   • Evaluated using Accuracy, Precision, Recall, and F1 Score for each traffic class.
   • Generated a confusion matrix to highlight misclassifications.

---

🤖 B. Unsupervised Learning with K-Means Clustering

1. 📏 Dataset Standardization:

   • Used StandardScaler to standardize feature values.

2. 🛠 Clustering Initialization:

   • Explored multiple cluster numbers (k) to uncover the data structure.
   • Used random_state for reproducibility.

3. 📉 Optimal Clusters:

   • Applied the Elbow Method to find the ideal k.

4. 🔗 Cluster Formation:

   • Grouped data into clusters and assigned labels.

5. ✔️ Cluster Validation:

   • Evaluated clustering with the Silhouette Score.

6. 🌈 Visualization:

   • Reduced dimensions using Principal Component Analysis (PCA) for 2D scatter plots.

---

📊 Results and Analysis

✅ Supervised Learning with Logistic Regression

• Confusion Matrix Highlights:
  • 🟢 Ping Traffic: 569 correctly classified.
  • 🟢 Voice Traffic: 885 correctly classified (highest accuracy).
  • 🟢 DNS Traffic: 574 correct, minor confusion observed.
  • 🟢 Telnet Traffic: 571 correctly identified samples.
  • 🟡 Misclassifications: Minimal errors, primarily between closely related traffic types (e.g., Telnet and DNS).

📊 Unsupervised Learning with K-Means

• Optimal clusters determined using the Elbow Method.
• Silhouette Scores indicated strong cluster separation.
• PCA Visualization: Clearly illustrated data separation.

---

🛠 Technologies and Tools Used

• 🐍 Python: Core programming language.
• 📚 Libraries: NumPy, Pandas, Scikit-learn, Matplotlib, Seaborn, TensorFlow.
• 🖧 SDN Simulation: Mininet and OpenFlow.
• 🔍 Dimensionality Reduction: PCA.

---

📚 Links

• Latex Report: link
• Project Explanation Video: link

---

📚 References

• Research Paper: Traffic Classification and Prediction in Computer Networks Using Machine Learning