M.Sc. Data Science Portfolio Project | Arden University
End-to-end machine learning pipeline predicting Indian domestic flight prices using statistical inference and ensemble methods.
The aviation sector operates as a volatile marketplace where fares shift dynamically with booking timelines, route coordination, and demand cycles. This project examines India's domestic flight pricing using statistical inference and predictive algorithms to decode cost drivers and project fare behavior.
Key Results:
- Best Model: Random Forest Regressor
- Performance: R² = 0.9783 | RMSE = ₹3,378 | MAPE = 10.56%
- Top Insight: Travel class alone drives 83% of price variation (Business class is ~698% more expensive than Economy)
- Source: Indian domestic flight dataset (
Flight_dataset.xlsx) - Size: 300,153 records × 12 features
- Target Variable: Ticket price (INR)
- Features: Airline, source city, destination city, departure/arrival time, stops, duration, days left, class
- Carriers: SpiceJet, Vistara, Air India, IndiGo, AirAsia, GO_FIRST
| Phase | Description |
|---|---|
| 1. Data Audit | Loaded raw data, identified missing values (22), duplicates (2,212), and schema issues |
| 2. Preprocessing | Removed identifiers, standardized text, imputed missing values, capped outliers (IQR method) |
| 3. Feature Engineering | Created route (30 directional pairs) and duration_bucket (4 time bands) |
| 4. EDA | Distribution analysis, correlation heatmaps, airline/class/route price breakdowns |
| 5. Hypothesis Testing | Welch t-test, ANOVA, Chi-square — all significant at p < 0.001 |
| 6. Modeling | Trained 5 regressors: Linear, Ridge, Lasso, Decision Tree, Random Forest |
| 7. Evaluation | Compared R², RMSE, MAE, MAPE; analyzed residuals and feature importance |
| Model | R² | RMSE (INR) | MAE (INR) | MAPE |
|---|---|---|---|---|
| Random Forest | 0.9783 | 3,378.30 | 1,645.90 | 10.56% |
| Decision Tree | 0.9681 | 4,094.36 | 2,109.96 | 13.41% |
| Ridge Regression | 0.8836 | 7,817.38 | 4,621.28 | 25.95% |
| Linear Regression | 0.8836 | 7,817.86 | 4,621.47 | 25.95% |
| Lasso Regression | 0.7346 | 11,805.04 | 7,102.17 | 42.08% |
- Class Premium Effect: Business class mean (₹52,591) is 697.95% above economy (₹6,591)
- Booking Horizon: Negative correlation (-0.09) between days_left and price — late bookings cost more
- Airline Hierarchy: Vistara (₹30,495 avg) and Air India (₹23,508) form the premium tier; AirAsia (₹4,099) is budget
- Route Effects: Chennai-Bangalore and Kolkata-Chennai routes carry higher average fares
- Feature Importance:
class_economy(83%),days_left(7%),duration_minutes(4.4%)
- Language: Python 3.x
- Libraries: pandas, numpy, scikit-learn, seaborn, matplotlib, scipy
- Environment: Jupyter Notebook / VS Code
- Reproducibility: Fixed random seed (SEED = 42)
# 1. Clone the repository
git clone https://github.com/DATA-Meta/ML-_arden_university.git
cd ML-_arden_university
# 2. Install dependencies
pip install -r requirements.txt
# 3. Launch Jupyter and open notebooks in order
jupyter notebookRun notebooks sequentially (01 → 02 → 03 → 04) for full reproducibility.
├── README.md
├── requirements.txt
├── data/
│ └── Flight_dataset.xlsx
├── notebooks/
│ ├── 01_data_cleaning.ipynb
│ ├── 02_eda_and_visualization.ipynb
│ ├── 03_hypothesis_testing.ipynb
│ └── 04_model_training.ipynb
├── reports/
│ └── STU234608_ML_Portfolio.pdf
└── .gitignore
Student ID: STU234608
Programme: M.Sc. Data Science, Arden University
Module: Machine Learning
Tutor: Mohammad Amin Mohammadi Banadaki
- James, G. et al. (2023) An Introduction to Statistical Learning: with Applications in Python. 2nd edn. Springer.
- Géron, A. (2022) Hands-On Machine Learning with Scikit-Learn, Keras and TensorFlow. 3rd edn. O'Reilly.
- Molnar, C. (2022) Interpretable Machine Learning. 2nd edn.
- McKinney, W. (2022) Python for Data Analysis. 3rd edn. O'Reilly.
- scikit-learn Developers (2024) scikit-learn: Machine Learning in Python.
This project was completed as part of academic coursework. Not intended for commercial deployment without further validation.