Airline-Booking

This project is a comprehensive Data Science and AI pipeline developed in three milestones, progressing from traditional Machine Learning prediction to Knowledge Graph construction, and finally to an end-to-end Graph-RAG (Retrieval-Augmented Generation) application for an Airline Travel Assistant.

Prerequisites

• Neo4j: Install Neo4j Desktop or Community Server locally.
• Python Libraries:

  pip install neo4j

---

MS 1: Airline Customer Holiday Booking (Predictive Modeling)

Goal: Develop a machine learning pipeline to predict passenger satisfaction based on customer feedback and flight data.

Key Features:

• Data Engineering: Cleaning datasets and performing Sentiment Analysis on review content using the Vader library.
• Exploratory Data Analysis (EDA): Visualizing top flight routes, booking distributions, and rating patterns by traveler type.
• Predictive Modeling: Implementing a Statistical ML model or Shallow Feed-Forward Neural Network (FFNN) to classify passengers as "Satisfied" (Rating ≥ 5) or "Dissatisfied".
• Model Explainability (XAI): Using SHAP and LIME to interpret model predictions and identify influential features.

Deliverables:

• Jupyter Notebook with reproducible workflow.
• Analytical Report & Visualizations.
• Predictive Model & XAI plots.

---

MS 2: Knowledge Graph Construction (Neo4j)

Goal: Transition from tabular data to a structured Neo4j Knowledge Graph (KG) that models relationships between passengers, flights, airports, and journeys.

Schema:

• Nodes: Passenger, Journey, Flight, Airport.
• Relationships: (:Passenger)-[:TOOK]->(:Journey), (:Journey)-[:ON]->(:Flight), (:Flight)-[:DEPARTS_FROM]->(:Airport), (:Flight)-[:ARRIVES_AT]->(:Airport) .

Key Tasks:

• KG Construction: A Python script (Create_kg.py) to ingest CSV data and build the graph adhering to a strict schema.
• Scoring Rule Implementation: Calculating a weighted overall_satisfaction_score based on food, delay, legs, and miles to identify passengers above a specific threshold.
• Cypher Analytics: Developing queries to answer business questions (e.g., busiest routes, average delays, food satisfaction by generation).

---

MS 3: End-to-End Graph-RAG Travel Assistant

Goal: Build a Graph Retrieval-Augmented Generation (Graph-RAG) system that uses the Neo4j Knowledge Graph from Milestone 2 as a grounding mechanism for an LLM-based assistant.

System Architecture:

1. Input Preprocessing:
   • Intent Classification: Routing queries (e.g., "Search" vs. "Recommend").
   • Entity Extraction: Using NER to identify airports, flights, and dates from user input.
2. Graph Retrieval Layer:
   • Baseline: Structured Cypher templates populated with extracted entities.
   • Embeddings: Semantic similarity search using vector embeddings (Node or Feature embeddings).
3. LLM Layer:
   • Combines retrieved graph context with a structured prompt (Context, Persona, Task) to generate accurate answers.
   • Comparison of at least 3 different LLMs (e.g., Llama, Gemma, Mistral).
4. User Interface:
   • A Streamlit UI to visualize the retrieved KG context, executed Cypher queries, and the final LLM response.

Install requirements

cd ms3
pip install -r requirements.txt