✈️ SkyRate — Airline Passenger Satisfaction Intelligence

🎓 Capstone Project — PGP Data Science & Generative AI | Great Learning | 2026

---

🌐 Live Demo

👉 Launch SkyRate App

📌 Project Overview

SkyRate is a full-stack AI-powered airline passenger satisfaction prediction system. It combines a conversational chatbot for passenger feedback collection with a real-time analytics dashboard for airline operations teams.

The system uses a 2-Stage Machine Learning architecture — Stage 1 predicts satisfaction from pre-flight experience, while Stage 2 analyzes in-flight service quality — both powered by XGBoost achieving 90-92% accuracy.

💡 Business Problem: Airlines struggle to identify dissatisfied passengers in real-time. SkyRate solves this by predicting satisfaction immediately after pre-flight check-in, enabling proactive service recovery before the flight departs.

---

🎯 Key Features

Feature	Description
🤖 AI Chatbot	7-step conversational feedback collection
📊 Live Dashboard	Real-time KPIs, charts, and passenger records
🔮 Instant Prediction	XGBoost predicts satisfaction in milliseconds
🎨 Custom Dark UI	Professional airline-grade interface
📥 CSV Export	Download all passenger data instantly
🔍 Smart Filtering	Search and filter by PNR, seat, satisfaction

---

🧠 ML Architecture — 2-Stage Model

┌─────────────────────────────────────────────────────┐
│                  STAGE 1 (Pre-Flight)                │
│  Demographics + Booking + Airport Experience         │
│  → XGBoost → 90.3% Accuracy | 92.7% Recall         │
└──────────────────────┬──────────────────────────────┘
                       │
┌──────────────────────▼──────────────────────────────┐
│                  STAGE 2 (In-Flight)                 │
│  WiFi + Entertainment + Comfort + Food + Crew        │
│  → XGBoost → 92.0% Accuracy | 93.7% Recall         │
└─────────────────────────────────────────────────────┘

Why 2 Stages?

• Stage 1 runs at check-in — early warning system for ground staff
• Stage 2 runs post-flight — identifies in-flight service gaps
• Prevents data leakage between pre-flight and in-flight features
• Enables targeted intervention at each journey stage

---

📊 Model Performance

Stage 1 — Pre-Flight Prediction

Model	Accuracy	Recall
Logistic Regression	83%	80%
Decision Tree	86%	84%
Random Forest	89%	88%
SVM	82%	79%
AdaBoost	88%	87%
XGBoost ✅	90.3%	92.7%

Stage 2 — In-Flight Prediction

Model	Accuracy	Recall
Logistic Regression	85%	82%
Decision Tree	88%	86%
Random Forest	91%	90%
SVM	84%	81%
AdaBoost	89%	88%
XGBoost ✅	92.0%	93.7%

📌 Why Recall over Accuracy? Missing a dissatisfied passenger costs the airline more than a false alarm. Recall ensures maximum detection of unhappy customers.

---

🔑 Key Insights

Stage 1 — Pre-Flight Drivers:
🏆 Online Boarding — strongest predictor of satisfaction
✈️  Type of Travel (Business) — business travelers demand higher service
📱 Ease of Online Booking — digital experience sets expectations

Stage 2 — In-Flight Drivers:
📶 In-flight WiFi (29%) — #1 satisfaction driver
🎬 In-flight Entertainment (22%) — critical for long-haul
💺 Leg Room Service (11%) — physical comfort matters

---

🛠️ Tech Stack

Layer	Technology
ML Models	XGBoost, Scikit-learn, Pipeline
Explainability	SHAP, Feature Importance
Clustering	KMeans + PCA
Frontend	Streamlit + Custom CSS
Charts	Plotly
Data	Pandas, NumPy
Deployment	Streamlit Cloud + GitHub Codespaces

---

🔄 How It Works

Passenger → SkyBot Chatbot (7 steps)
                ↓
    Ratings collected (booking, check-in, boarding, gate)
                ↓
    Stage-1 XGBoost predicts satisfaction (hidden from passenger)
                ↓
    Result stored → Dashboard updates instantly
                ↓
    Airline team sees: KPIs + Charts + Passenger Table

---

📁 Project Structure

SkyRate-Airline-Customer-Satisfaction/
│
├── 🐍 app.py                    ← Main entry point (tabs + CSS)
├── 🤖 model.py                  ← XGBoost Stage-1 pipeline
├── 📋 requirements.txt          ← Dependencies
├── 🚫 .gitignore
│
├── 📂 pages/
│   ├── 💬 chatbot.py            ← 7-step passenger feedback bot
│   └── 📊 dashboard.py          ← Live airline analytics dashboard
│
├── 📂 .devcontainer/
│   └── devcontainer.json        ← GitHub Codespaces config
│
├── 📓 Stage1_Notebook.ipynb     ← Pre-flight ML analysis
├── 📓 Stage2_Notebook.ipynb     ← In-flight ML analysis
│
└── 📄 README.md

🧠 What I Learned

• Building end-to-end ML pipelines with no data leakage
• 2-Stage model architecture for sequential prediction problems
• SHAP explainability for model interpretability
• KMeans + PCA for customer segmentation
• Deploying ML models as production Streamlit apps
• Custom UI design with CSS in Streamlit
• Real-time state management in multi-page apps

---

👨‍💻 Author

Kiran U

BCA Graduate | PGP in Data Science & Generative AI — Great Learning, Bangalore

⭐ If you found this project useful, please give it a star!