In bike-sharing systems, a small percentage of the bicycles become unusable every day. Currently, there is no reliable on-line information that indicates the usability of bicycles. We present a model that estimates the probability that a specific bicycle is unusable as well as the number of unusable bicycles in a station, based on available trip transaction data. Further on, we present some information based enhancements of the model and discuss an equivalent model for detecting locker failures.
- Bayesian model