Project

Bike Share Demand Forecasting Space–Time Prediction + Rebalancing Insight · Capital Bikeshare (Washington, DC)

What this study does

This project builds a demand forecasting workflow for bike share systems using time-based patterns, weather signals, and lagged ridership to predict near-term trip volume and station pressure. The goal is operational: support smarter rebalancing by anticipating when and where demand spikes.

Focus

Demand forecasting + operational rebalancing

Methods

Space–time features, lag variables, model evaluation

GIS

Spatial joins + neighborhood context variables

Tools

R Studio · (GIS/Spatial workflow)

View full technical report

Observed vs predicted demand: model performance overview — Model performance overview: observed vs predicted demand across the study period.

Key outputs

Trips over time time series — Ridership time series capturing baseline demand and volatility.

Weather relationship plot — Weather sensitivity: how conditions relate to ridership changes.

Errors by model comparison — Model comparison using error metrics to select the best-performing approach.

Station-level errors — Station-level error patterns highlight where predictions are most uncertain.

Problem

Bike share systems fail when demand is misaligned with supply: stations become empty (no bikes) or full (no docks). Forecasting demand helps operators rebalance proactively instead of reacting after failures occur.

Approach

Engineer time features (hour/day patterns) and lagged demand signals.
Integrate weather and calendar effects to capture external demand drivers.
Evaluate performance using observed vs predicted comparisons and station-level error behavior.

GIS + Spatial Context

The workflow incorporates spatial context by associating ridership/station dynamics with neighborhood-level conditions (e.g., access, surrounding land-use or socio-economic proxies) to interpret where forecasting is harder and where operations are most sensitive.

What this demonstrates

Technical depth: feature engineering, modeling, evaluation, and interpretation.
Modeling: time-sensitive prediction and error diagnostics.
GIS: spatial joins and place-based analysis to contextualize operational outcomes.

Open full technical report

Station summary figure — Station summary: ridership distribution and station behavior used to interpret operational pressure.