CitiBike Inventory Prediction

A machine learning project for predicting bike station inventory at NYC CitiBike stations using time-series forecasting and Markov chain models.

Overview

This project implements multiple models to predict how many bikes will be available at each CitiBike station over time:

  • PersistenceModel: Baseline that assumes inventory stays constant

  • StationAverageModel: Uses average net flow per station

  • TemporalFlowModel: Time-conditioned flow predictions

  • MarkovModel: Markov chain with transition matrices and Monte Carlo simulation

Features

  • Rolling window cross-validation for time-series evaluation

  • Multiple evaluation metrics (MAE, RMSE, state classification)

  • Configurable via YAML configuration files

  • Reproducible random seeds for stochastic models

Indices and tables