Table of Contents
Last update: January 2024. All opinions are my own.
1. Overview
This project predicts hourly bike usage for Washington, D.C. It covers exploratory analysis, feature engineering, and time-based validation to avoid leakage.
2. Data and Features
The dataset includes calendar and weather signals that drive demand. Capturing daily cycles and seasonal trends is essential for good forecasts.
- Engineered time-based features for hourly and weekly patterns.
- Explored weather effects on usage spikes and drop-offs.
- Reviewed distributions to guide transformations.
3. Modeling Approach
- Established baseline forecasts for comparison.
- Evaluated models using time-based splits.
- Iterated on features and validation strategy.
4. Insights
The biggest lift came from feature engineering around time and weather, plus strict time-aware validation.
5. Skills and Tools
- Python
- Data visualization
- Time series validation