A cluster-driven classification approach to truck stop location identification using passive GPS data

Classifying the type of truck stops is vital in transportation planning and goods movement strategies. Truck stops could be classified into primary or secondary. While the latter entail stopping to re-fuel or rest, the former takes place to deliver or pick up merchandize. The availability of GPS transponders on board moving trucks and the ability to access such information in recent years has made it possible to analyze various freight aspects including movement trajectories and stopped locations. This paper utilizes machine learning methods and proposes a two-step cluster-based classification approach to classify truck stop locations into either primary or secondary. The DBSCAN clustering technique is applied on the GPS dataset to obtain stop locations. Next, several features per location are derived to classify the stops using well-known classification models. The generated information is then used to evaluate the approach using a large truck GPS dataset for the year 2016. The Random Forest classifier is chosen as it can identify primary stop locations with an accuracy of 97%. The overall accuracy of the classifier for correctly identifying both types of stops is 83%. Further, the prediction accuracy for primary stops is 92%.