Comparison of Spatial Predictability Differences in Truck Activity Patterns: An Empirical Study Based on Truck Tracking Dataset of China

Existing research on truck location prediction focuses on direct trajectory prediction and ignores the link between activity patterns and predictability, whereas the mode of operation is an important factor in the difference between activity trajectories, and analyzing the mode of operation can help to develop the next-location prediction algorithms to improve the efficiency of matching truckloads and to reduce costs. Our empirical study, based on 562,071 truck trip data in China, employs Fuzzy c-means (FCM) for clustering operational patterns in space, intensity, and stability dimensions. K-nearest neighbors (KNN), Back Propagation neural (BP) network, and Long Short-Term Memory (LSTM) predict the next truck locations in different modes. The results indicate that range-of-motion stability significantly influences predictability. Truckers with stable spatial activity exhibit the highest predictability, with 45% nearly achieving 100% predictability. Through cluster analysis of driving characteristics, we found that truck clusters are the most predictable because of their relatively small and low-intensity activities, with the percentage of samples with prediction accuracy above 90% reaching over 80%. This research not only characterizes the freight truck community but also aids algorithm optimization by revealing predictability factors for real-world applications.