No abstract

No abstract

Smart card data of conventional bus passengers are important basic data for many studies such as bus network optimization. As only boarding information is recorded in most cities, alighting stops need to be identified. The classical trip chain method can only detect destinations of passengers who have trip cycles. However, the rest of unlinked trips without destinations are hard to analyze. To improve the accuracy of existing methods for determining alighting stops of unlinked trips, a two-layer stacking-framework-based method is proposed in this work. In the first layer, five methods are used, i.e., high-frequency stop method, stop attraction method, transfer convenience method, land-use type attraction method, and improved group historical set method (I-GHSM). Among them, the last one is presented here to cluster records with similar behavior patterns into a group more accurately. In the second layer, the logistic regression model is selected to get the appropriate weight of each method in the former layer for different datasets, which brings the generalization ability. Taking data from Xiamen BRT Line Kuai 1 as an example, I-GHSM given in the first layer has proved to be necessary and effective. Besides, the two-layer stacking-framework-based method can detect all destinations of unlinked trips with an accuracy of 51.88%, and this accuracy is higher than that of comparison methods, i.e., the two-step algorithms with KNN (k-nearest neighbor), Decision Tree or Random Forest, and a step-by-step method. Results indicate that the framework-based method presented has high accuracy in identifying all alighting stops of unlinked trips.

No abstract