Spatial forecasting of online food delivery demand

AbstractThis work aims to predict demand for orders in the expanded center of S˜ao Paulo, Brazil, using the Spatial Autoregressive Model (SAR) and the Artificial Neural Network - Multilayer Perceptron Model (MLP), based on real data from a Brazilian online food delivery company. Accurately forecasting demand by geographical area is crucial for efficient logistics planning in the company. To address this, we proposed an approach that uses an augmented train matrix to incorporate order information from neighboring areas of the first order, allowing the neural network model to identify spatial autocorrelation in the data. The performance of both models was evaluated using Root Mean Squared Error (RMSE) and the coefficient of determination (R2). In the simulations, both of the models achieved satisfactory performance compared to the historical mean. The MLP model, when best tuned, outperformed the SAR model in both RMSE and R2coefficient.