Traffic Demand Prediction

Machine learning pipeline for traffic demand forecasting using CatBoost and spatio-temporal feature engineering.

Team Project

This repository contains:

• Dataset files
• Final submission file
• Complete training pipeline
• Feature engineering workflow

Features Engineered

• Geohash encoding
• Time-slot segmentation
• Cyclical time encoding
• Day-48 historical lookup feature
• Target encoding features
• Road/weather aggregation features

Model Used

• CatBoost Regressor

Validation Performance

• R² Score: ~0.9926
• Approx competition score: ~99.26

Important Insight

The strongest predictive feature was:

same geohash + same time slot historical demand

showing strong spatio-temporal repetition patterns in traffic demand.

Files

train.csv
test.csv
sample_submission.csv
submission.csv
traffic_demand_solution.py

Run

pip3 install pandas numpy scikit-learn catboost
python3 traffic_demand_solution.py