doluMu
BSc thesis — passenger crowding forecasting for Istanbul public transport
- →Built a forecasting system that predicts hourly passenger intensity for Istanbul bus and rail lines over the next 24 hours
- →Converted passenger count forecasts into occupancy metrics by combining predictions with route capacity/trip information
- →Delivered the product as a public, login-free PWA
Data & Modeling
- ·Cleaned IBB hourly journey data and aggregated it to route-hour level
- ·Enriched data with calendar signals and hourly weather data from Open-Meteo
- ·Engineered lag features (24/48/168), rolling statistics, and time-based features
- ·Trained a global LightGBM model that learns across hundreds of routes
- ·Tracked and interpreted model behavior with MLflow and SHAP
Product & Deploy
- ·Scheduled nightly batch forecasting — wrote predictions to database tables
- ·Exposed ready predictions via FastAPI API
- ·Ran FastAPI + PostgreSQL in Docker on a remote server behind Nginx with HTTPS
- ·Built frontend in Next.js with PWA support, caching, and local storage