Fig 1.
Diagram summarizing our bicycle implicit route choice model.
Fig 2.
Distribution of distance differences between the experienced tracks and the generated shortest paths (sorted).
Fig 3.
Heat-maps of the experienced tracks (on the left) and the computed tracks (on the right).
Fig 4.
CDF of the differences of distances between experienced and computed tracks.
Fig 5.
Graphical example of the intersection between the extended sets of cells of two tracks.
Fig 6.
Evolution of the noise (green line), the number of clusters (red line), the number of large clusters (black line) and the mean coefficient silhouette (blue line) with respect to ε and MinPts = 3.
Fig 7.
Heat-maps representing the experienced tracks contained in the largest cluster found by DBSCAN in Lyon (left) and their equivalent computed tracks (right).
Fig 8.
Box-plots of the distance between the experienced tracks and the four other types of tracks.
Fig 9.
Diagram showing the structure of a LSTM cell.
Fig 10.
Successed and failed predictions of our LSTM in each cluster in the testing data-set.
Fig 11.
Boxplots of the distance between the experienced tracks and the experience-based tracks.
The left one shows the testing data-set and the right one the training data-set.
Fig 12.
Maps showing four experienced tracks (orange lines), the corresponding computed tracks (blue lines) and the experience-based tracks created with our model (red lines) between the experienced tracks’ O/D pairs.