Network Traffic Congestion Predictor

A machine learning solution for proactive network management

---

Overview

This project builds a system that analyzes network traffic patterns and predicts whether congestion is likely to occur, giving network administrators a heads-up to take action.

The Problem: Network congestion leads to slow load times, dropped connections, and frustrated users.

The Solution: By predicting congestion before it happens, ISPs and network teams can implement traffic shaping, load balancing, or capacity adjustments proactively.

---

Quick Start

# Clone and navigate
git clone https://github.com/AR10129/Network-Traffic-Congestion-Prediction.git
cd Network-Traffic-Congestion-Prediction

# Set up environment (Windows)
python -m venv .venv
.venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Train the model
python scripts/run_pipeline.py

# Launch the web app
python app.py

Then visit http://127.0.0.1:5000

---

Project Structure

├── app.py             Flask web application
├── requirements.txt   Python dependencies
├── config/            YAML configuration files
├── data/              Generated traffic datasets
├── logs/              Application logs
├── models/            Trained model artifacts (.pkl)
├── notebooks/         Jupyter notebooks for EDA
├── screenshots/       App demo images
├── scripts/           Pipeline execution scripts
├── src/               Core source code modules
│   ├── data/              Data generation utilities
│   ├── features/          Feature engineering logic
│   ├── models/            Training & prediction code
│   ├── utils/             Logging & helper functions
│   └── visualization/     Plotting functions
├── static/            CSS stylesheets
├── templates/         HTML templates for Flask
└── visualization/     Generated plots and charts

---

Screenshots

Homepage	Prediction Form	Result

---

How It Works

ML Pipeline

Generate Data  -->  Engineer Features  -->  Train Model  -->  Deploy via API

Stage	Description
Data Generation	Synthetic network traffic with realistic peak/off-peak patterns
Feature Engineering	IP conversion, packet rate calculation, temporal features, one-hot encoding
Model Training	Random Forest with TimeSeriesSplit CV + RandomizedSearchCV hyperparameter tuning
Deployment	Flask web app for real-time predictions

Key Features Extracted

Feature	Description
packet_size	Size of network packets
bytes_sent	Total bytes transmitted
packet_rate	Rolling average of packet transmission
hour / day_of_week	Temporal patterns
is_weekend	Weekend indicator (different traffic patterns)
protocol_*	One-hot encoded protocol type (TCP/UDP/HTTP)

---

Model Performance

The model uses Random Forest Classifier with the following evaluation approach:

• Cross-Validation: TimeSeriesSplit (prevents data leakage in temporal data)
• Hyperparameter Search: RandomizedSearchCV for efficient tuning
• Metrics Tracked: Precision, Recall, F1-Score, AUC-ROC

After training, visualizations are auto-generated:

• Confusion Matrix
• Feature Importance Chart
• Traffic Volume Over Time
• Protocol Distribution

---

Web Interface

The Flask app provides a simple interface to test predictions:

Input Fields:

• Packet Size
• Bytes Sent
• Source/Destination IP
• Protocol (TCP/UDP/HTTP)
• Timestamp & Hour

Output: Congested or Normal prediction

---

Configuration

All parameters are centralized in config/config.yaml:

# Model hyperparameters
model:
  param_grid:
    n_estimators: [50, 100, 200, 300]
    max_depth: [null, 10, 20, 30]
  cv_splits: 5

# Data settings
data:
  test_size: 0.2
  random_state: 42

---

Tech Stack

Category	Technologies
ML/Data	pandas, numpy, scikit-learn
Visualization	matplotlib, seaborn
Web Framework	Flask
Config	PyYAML
Utilities	ipaddress, pickle

---

Roadmap

• Integrate real-time network monitoring (SNMP/NetFlow)
• Experiment with Gradient Boosting / XGBoost
• Add model drift detection
• Docker containerization
• REST API with authentication
• Dashboard with live metrics

---