OnTime

The idea for this project came from my own frustration with flight delays. I once stood at the airport watching my flight go from “On Time” to “15 minutes delayed,” then “30 minutes,” then “1 hour.” By the time I realized how bad it was, I had already missed an important family event. That moment made me wonder: what if we could know delays earlier and make better choices?

That’s why I built OnTime, a global flight delay prediction system. It collects live data from 60+ airports and uses machine learning — Linear Regression, Ridge, Lasso, and Random Forest — to predict delays in minutes. We engineered 20+ features like airline, route, time of day, and past delay patterns, then chose the best-performing model with metrics like R^2 and RMSE. The predictions are shown in clear buckets (+15m, +30m, +45m) through a modern, easy-to-use interface.

The hardest part was dealing with data. Real flight APIs are expensive or limited, so we had to combine scraping with realistic synthetic generation. We also had to make all the pieces work together: scrapers, database, machine learning, backend, and frontend. Through this, I learned how to connect full-stack engineering with real-world machine learning.

In the end, OnTime is more than just code. It’s about turning the stress of flying into confidence, giving travelers foresight instead of frustration.