Airline Passenger Satisfaction — Data Mining Project

This repository contains an end-to-end data mining project built around a simple goal: understand what drives airline passenger satisfaction and predict it reliably.
Along with model training and evaluation, a working React and Flask web app is included so predictions and insights can be explored through a clean dashboard instead of only notebooks.

---

Project Demo (YouTube)

https://youtu.be/TPWopQruvhI

---

GitHub Repository

https://github.com/ttabirami12062/DataMining-AirlineSatisfaction

---

What this project does

This project supports three useful outcomes:

1) Predict satisfaction for an individual passenger
Given travel + service details, the system returns:

• a Satisfaction label (Satisfied / Not Satisfied)
• a Satisfaction score (0–1 style score from a regression model)
• a Passenger segment (cluster assignment)

2) Explain the main drivers behind satisfaction
Regression models are used to make the relationship between features and satisfaction easier to interpret (service scores, delays, and travel context).

3) Segment passengers into groups
PCA + K-Means are used to group passengers into clusters that represent different experience patterns (ex: high-service vs low-service groups).

---

Methods Included

Regression (scoring + interpretability)

Used to produce a satisfaction score and understand feature impact:

• OLS
• Ridge
• Lasso
• Piecewise regression (where applicable)

Clustering (passenger segmentation)

• PCA for dimensionality reduction
• K-Means for clustering
• Cluster quality checked using standard metrics (Elbow / Silhouette / CH score depending on the experiment design)

Classification (main prediction)

• Random Forest (implemented from scratch in the project workflow)
• Tuned to improve generalization and reduce overfitting

---

Dataset

Dataset source: Kaggle – Airline Passenger Satisfaction
Typical fields include passenger type, travel type/class, delays, and multiple service rating categories.
The target is converted into a binary label:

• Satisfied = 1
• Neutral/Dissatisfied = 0

---

Data Preparation (high-level)

The dataset is prepared for modeling by:

• removing non-informative ID columns
• handling missing values (median for numeric, mode for categorical)
• encoding categorical variables (binary mapping + one-hot encoding where needed)
• engineering practical features such as:
  • Total_Delay = Departure Delay + Arrival Delay
  • Total_Service_Score = average of service rating features

---

Web App Overview (React + Flask)

Frontend (React)

• Pages commonly include: Home, About, Predict, Reports, Dashboard
• Predict page allows user input and shows model outputs clearly

Backend (Flask)

• Loads trained models (.pkl)
• Recreates engineered features during inference
• Exposes endpoints for predictions and dashboard summaries

Example endpoints:

• POST /api/predict
• GET /api/dashboard-summary