Achieved 15th Place (out of 578) in the Kaggle FlightRank 2025: Aeroclub RecSys Cup.
In the complex world of business travel, choosing the right flight isn't just about the lowest price—it's about balancing cost, time, and comfort. This engine automates that choice by:
- Optimizing Travel Budgets: Identifying the most cost-effective routes without sacrificing quality.
- Reducing Search Fatigue: Ranking thousands of options to show the Top-3 most relevant flights instantly.
- Smart Personalization: Learning hidden patterns in traveler behavior using Deep Learning.
The solution is built using a sophisticated multi-model architecture:
- Deep & Cross Network (DCNv2): A state-of-the-art neural network architecture that explicitly learns feature interactions, perfect for complex tabular data like flight attributes.
- Learning to Rank (LTR): Direct optimization of the NDCG metric to ensure the most relevant flights appear at the very top of the search results.
- Geospatial Intelligence: Dynamic Haversine distance calculations between 1,000+ global airports to evaluate route convenience.
The repository is structured as a professional data science production pipeline:
- Data Sanitization: Cleaning raw datasets and handling missing values.
- Feature Extraction: Deep parsing of nested JSON objects to find hidden signals.
- Geo-mapping: Integrating airport coordinates for distance-based ranking features.
The production cycle includes our 4 high-performing models and the final ensemble:
- Neural Models: DCNv2 and Stacked MLP architectures.
- Gradient Boosting: Specialized rankers using CatBoost and XGBoost (the industry gold standard).
- Final Ensemble: A weighted rank-blending strategy that combines the strengths of all models to achieve maximum precision.
- Install dependencies:
pip install -r requirements.txt - Prepare data: Run notebooks in
01_preprocessingsequentially. - Generate results: Execute the final blending script in
notebooks/02_production/05_Final_Ensemble_Blending.ipynb.
Developed by Nikolay Alymov Expertise in AI Automation, RecSys, and Enterprise Machine Learning.