Traffic Demand Prediction - HackerEarth

Solution for the Traffic Demand Prediction hackathon challenge.

Problem Statement

Predict traffic demand at different geo-locations based on road type, weather, time, and other features.

Evaluation: max(0, 100 * R2_score(actual, predicted))

Approach

Feature Engineering

• Parsed timestamp (H:M) into hour and minute
• Added cyclical features (sin/cos) for time of day
• Encoded categorical features: RoadType, Weather, LargeVehicles, Landmarks
• Label encoded geohash (1,249 unique locations)
• Filled missing Temperature with median
• Created interaction features: road_lanes, is_rush_hour, is_night

Model

• LightGBM Regressor
• 1500 estimators, learning_rate=0.03
• 5-Fold Cross Validation for evaluation
• Predictions clipped to [0, 1]

Files

• traffic_demand_solution.py - Main Python solution
• traffic_demand_prediction.ipynb - Detailed notebook with EDA
• submission.csv - Final predictions

How to Run

pip install pandas numpy lightgbm scikit-learn
python traffic_demand_solution.py