How to Predict Scooter Demand? XGBoost and Urban Micromobility

XGBoost on 341 temporal, spatial and network features predicts e-scooter demand 27-49% better than rivals and up to 90% better than Prophet, node centrality on top.

Przeczytaj ten artykuł po polsku

Can we predict when and where people will rent electric scooters? Yes — and with impressive accuracy. A recent publication shows how advanced algorithms like XGBoost can revolutionize the management of micromobility in cities.

🌍 Context: Micromobility and Demand

In many cities, dockless electric scooters have become a daily transport option. But for operators, a crucial question remains:

Where and when will people want to rent a scooter?

Too many vehicles in one location is wasteful. Too few — lost revenue and frustrated users. That’s why accurately predicting demand is so important.

🧠 The Paper: A Three-Dimensional Prediction Model

The authors of arXiv 2507.02715 proposed a comprehensive machine learning framework combining three types of data:

  • Temporal (e.g., hour, day of the week, weather),
  • Spatial (e.g., proximity to public transport, number of POIs nearby),
  • Network-based (e.g., how frequently users travel between different areas).

They generated 341 input features from this data and fed them into an XGBoost model.

🌳 What is XGBoost?

XGBoost stands for eXtreme Gradient Boosting. It’s a model that:

  • Consists of multiple decision trees,
  • Each subsequent tree corrects errors of the previous ones (gradient boosting),
  • Works very efficiently with tabular data.

🛠️ How does it work?

  1. Starts with a simple prediction (e.g., average),
  2. Calculates error and gradients,
  3. Builds a new tree to model those gradients,
  4. Adds the tree’s output to the previous prediction.

Final formula:

📈 Why does it work?

In the paper, the model performs well because:

  • The input features were well-engineered,
  • XGBoost handles non-linearity (e.g., temp, time of day),
  • SHAP analysis showed network topology (e.g., node centrality) were top features.

🔬 Results

  • The model improved accuracy by 27–49% compared to other methods,
  • Up to 90% better than classical time series models like Prophet.

👉 arXiv:2507.02715

Related articles

Comments

Comments are moderated before they appear.

  • Loading comments…