🛰️ Network Traffic Analyzer & Optimization System

A Streamlit-based real-time application for capturing, analyzing, classifying, and optimizing network traffic using machine learning models. The system leverages Scapy, KMeans, Decision Trees, and XGBoost to build a lightweight pipeline for monitoring bandwidth, jitter, and traffic types.

---

📌 Problem Statement

Modern networks require continuous monitoring to maintain Quality of Service (QoS) and prevent congestion. Low-level packet inspection is time‑consuming and noisy, making it difficult to identify actionable insights such as traffic type (VoIP, browsing, streaming) or resource bottlenecks.

This project solves that by:

• Automating packet capture
• Extracting statistical features
• Classifying traffic types
• Predicting bandwidth and jitter
• Suggesting optimization actions in real time

---

📄 Abstract

This tool captures packets in 10‑second windows and extracts nine interpretable features (burst rate, packet size, entropy, jitter, etc.). Using these features:

• KMeans clusters the traffic into 5 groups.
• Clusters are mapped to real-world classes (VoIP, Browsing, Streaming, Interactive TCP, Unknown).
• A DecisionTreeClassifier learns to classify future batches.
• Two XGBoost models predict bandwidth and jitter.

The Streamlit interface provides:

• Live metrics for every batch
• Automatic optimization actions
• Historical analytics (bandwidth trends, jitter trends, traffic distribution)

---

📊 Data Capture & Feature Extraction

Traffic data is collected from the local network using Scapy:

Each capture window is converted into a single feature vector:

Feature	Description
burst_rate	Packets per second
avg_packet_size	Mean size of packets
protocol_ratio_TCP	% of TCP packets
protocol_ratio_UDP	% of UDP packets
size_scaled	Min-max normalized packet size
port_entropy	Entropy of source ports
inter_packet_gap	Average time gap between packets
jitter	Std deviation of gaps
byte_ratio_uplink_downlink	Ratio of LAN uplink to downlink bytes

These features are used for modeling traffic behavior.

---

🔍 Literature‑Inspired Approach

• Clustering to infer traffic types from unlabeled data
• Tree‑based models for interpretability
• Regression for continuous QoS indicators

---

⚙️ Methods & Implementation

1️⃣ Phase 1 — Unsupervised Learning

• Capture multiple batches
• Extract features -Train KMeans and assign traffic types using simple heuristic rules -(Streaming, VoIP, Browsing, Interactive TCP, Unknown).
• Train models:
  • Decision Tree → Classify traffic type
  • XGBoost → Predict bandwidth (Mbps)
  • XGBoost → Predict jitter (ms)
• Save models to disk using Joblib

2️⃣ Phase 2 — Monitoring

• Load trained models
• Capture live batches
• Predict traffic type, bandwidth, jitter
• Trigger suggested optimizations
• Update dashboard and logs

---

✅ Main Results

• Real-time traffic classification

• Bandwidth & jitter prediction with ML regressors

• Automated action suggestions

• Detailed analytics:

  • Bandwidth trend chart
  • Jitter trend chart
  • Traffic distribution (pie)
  • Bandwidth per traffic type (bar)

• Lightweight statistical features can classify traffic effectively.

• KMeans + rule-based labeling is a strong alternative when labeled data is absent.

• Decision Trees offer clarity for network operators.

• Jitter and Bandwidth prediction is useful for stability monitoring.

• Real-time windowing gives actionable short-term QoS signals

---

🚀 Features

• Live capture using Scapy
• Dynamic Streamlit UI
• Auto-updating charts
• Automatic ML model training and loading
• Two operational phases:
  • Phase 1: Learning
  • Phase 2: Monitoring

---