Duration and Vehicle Utilization Forecasting for Car Sharing
Project Overview
This research addresses key operational challenges in the car-sharing industry by developing a forecasting framework for booking duration and vehicle utilization at the city level.
Leveraging real-world operational data from Co-Wheels, we implemented time series forecasting models—including Auto/Manual ETS, SARIMA, and Linear Regression—to generate 30-day forward predictions. These models were evaluated based on their accuracy and robustness under different demand scenarios.
Research Objectives
- Predict booking durations across cities to support dynamic fleet allocation.
- Estimate vehicle utilization using booking logs and real usage data.
- Improve forecasting performance through automated model selection, data decomposition, and seasonality analysis.
Methodology Highlights
- Data Processing: Outlier detection using Z-score; imputation via seasonal patterns; structural breakpoint identification.
- Modeling: Comparison between automatic and manual ETS/SARIMA; linear regression for interpretable trend analysis.
- Evaluation: Forecast accuracy assessed using MAE, MSE, and RMSE metrics across total and day-of-week slices.