Optimization of Bicycle Path Networks Using Autoregressive Models: An Application to the City of Querétaro

Many studies focus on the location of bicycle stations rather than network design, often overlooking the cyclist’s experience, which depends on individual characteristics and the type of bicycle used. This study proposes a methodology for designing bicycle corridors using GPS data and autoregressive models with exogenous variables (ARX). Two experiments were conducted: (1) a fixed-route scenario varying cyclists and bicycles, and (2) a variable-route scenario using the cyclists’ own bicycles. Both experiments evaluated the model’s accuracy and “transferability” using historical datasets to predict subsequent speeds. Results show an R-value of 0.94 for fixed routes and 0.93 for variable routes, with lags of 255 and 139 historical data points, respectively. The best-performing model for fixed routes achieved an MAE of 2.01 km/h, an RMSE of 2.55 km/h, a MAPE of 0.13%, and an R-value of 0.92 with 238 lags; with only 2 lags, the R-value was 0.82. Public and six-speed bicycles demonstrated greater transfer compatibility than electric ones. Finally, the model (using two lags) was applied to the road network in Querétaro to create a preliminary corridor design. The study concludes that while the model is useful—and could better represent cyclist behavior with additional data—it should not be viewed as a universal solution.