Trip Purpose Imputation Using GPS Trajectories with Machine Learning

We studied trip purpose imputation using data mining and machine learning techniques based on a dataset of GPS-based trajectories gathered in Switzerland. With large number labeled activities eight categories, we explored location information hierarchical clustering achieved classification accuracy 86.7% random forest approach as baseline. The contribution this study is summarized below. Firstly, from GPS exclusively without personal shows negligible decrease (0.9%), which indicates the good performance our steps wide applicability scheme case limited availability. Secondly, dependence model geographical location, participants, duration survey investigated to provide reference when comparing accuracy. Furthermore, show ensemble filter be an excellent tool research field not only because increased (93.6%), especially for minority classes, but also reduced uncertainties blindly trusting labeling by vulnerable class noise due response burden. Finally, derivation across participants reaches 74.8%, significant suggests possibility effectively applying trained small subset citizens larger trajectory sample.