FlightIQ

End-to-end model selection and validation pipeline, including benchmarking, hyperparameter optimization, and SHAP-based interpretability.
Structured model evaluation pipeline demonstrating why XGBoost was selected over linear and ensemble baselines, with cross-validation.
Cooperation with Teammates

Inspiration

Booking a flight usually comes down to price and departure time. But after a few real trips, we realized that what truly ruins the experience is reliability: routes that are consistently delayed, airports that act like bottlenecks, and “good-looking” options that are actually unstable because there isn’t enough flight volume behind them.

The data exists, but it’s fragmented across different tables and not presented in a decision-friendly way. We built FlightIQ to turn official reliability statistics into something users can compare in seconds.

What it does

FlightIQ is a web app that converts official BTS (Bureau of Transportation Statistics) on-time data into a clear, explainable Route Score (0–100). Users can:

Filter results by month, and optionally by origin and destination airports.
View ranked routes with a score and supporting metrics.
Explore airline on-time rankings and airport on-time rankings.
Click any route to see a transparent score breakdown (no black-box scoring).

How we built it

We designed FlightIQ as a full pipeline: data → scoring → interactive UI.

Data

We use official BTS on-time datasets as the primary source for route/airport/airline reliability signals.
This includes historical performance that supports ranking and comparison.

Scoring (Explainable by design)

We compute a route-level on-time signal from airport performance.
We then generate a composite score with a simple, interpretable formula:

[ \text{Route Score} = 0.8 \cdot \text{On-time Score} + 0.2 \cdot \text{Flight Volume Score} ]

The flight volume score is normalized to a 0–100 scale using percentiles under the current filters, so users can compare routes fairly within the same context.

Frontend Experience

The UI supports filtering, sorting, and route-level drill-down so users can understand both the score and the reasons behind it.

Challenges we ran into

Data joining and consistency: Official datasets are reliable, but not always “analysis-ready.” We had to standardize keys and ensure consistent route/time representations.
Fair comparisons: A route can look great with a small sample size. We addressed this by incorporating flight volume as part of the score.
Explainability vs. complexity: It’s easy to build a complicated model that nobody trusts. We intentionally kept the scoring logic simple and transparent so it can be inspected and justified.

Accomplishments that we're proud of

Built an end-to-end system that turns raw official statistics into an interactive reliability product.
Delivered a Route Score /100 that balances reliability and stability, not just “best-case” performance.
Implemented route drill-down explainability, showing score composition and metric contributions.
Added multi-level insights: route rankings, airline rankings, and airport rankings in one place.

What we learned

In real data products, having data is not the same as having usable data. Standardization and careful key design matter.
Users don’t trust black boxes. An explainable scoring system can be more valuable than a complex model if it supports clear decisions.
Good UI is part of the solution: ranking + filtering + drill-down makes the data actually usable.

What's next for FlightIQ

Add more context features (e.g., seasonal patterns, route direction effects) while keeping interpretability.
Improve performance for large result sets with caching and smarter server-side aggregation.
Expand the UI to support comparisons (e.g., compare two routes side-by-side) and shareable links for filtered views.