Bike-Share Demand Forecasting
Hourly bike-share rental demand forecasting with honest time-series validation, SHAP explainability, and a containerized FastAPI serving layer.
Problem
Bike-share operators need to predict hourly rental demand to rebalance bikes across stations, but naive forecasts ignore weather, time-of-day and seasonal effects.
Solution
An hourly demand-forecasting model with honest time-series validation and SHAP explainability, deployed behind a containerized FastAPI serving layer.
Overview
Bike-share systems live or die on rebalancing — bikes need to be moved to where riders actually are, which means operators need an hourly demand forecast, not just a daily one.
How it works
An hourly forecasting model incorporates weather, time-of-day and seasonal patterns, validated with honest time-series splits so its reported accuracy reflects real forecasting performance. SHAP explains which factors are driving each hour’s predicted demand, and the model is deployed behind a containerized FastAPI endpoint so it can be queried in real time.
Result
An hourly, explainable demand signal operators could actually use to plan bike rebalancing routes.
Further reading
See Why Most Data Portfolios Get Rejected and From Notebook to Data Product for more on honest evaluation splits and serving models through an API.
More projects
Visual Quality Inspection
Automated visual quality control for manufacturing, using OpenCV/scikit-image feature extraction with scikit-learn classification.
Fraud Detection System
A classical machine learning system for detecting fraudulent transactions with a focus on precision/recall trade-offs.
Marketing Attribution Warehouse
A marketing attribution data warehouse and modeling pipeline for multi-touch attribution reporting.