Application of BP Network for Travel Behavior Analysis: Complexity Recognition of Trip Chaining

The article develops a BP network for trip chaining pattern recognition based on the data obtained from Beijing Resident Trip Survey. First a set of socioeconomic and demographic factors related to traveller information which potentially influence trip-chaining patterns are pre-treated through principle components analysis, therefore seven variables are selected as input variables of neural network, and a categorical trip chaining pattern (simple and complex trip chaining) are used as output variables. In order to quantify prediction accuracy, two performance measures are applied to evaluate it. Besides, a logistic regression model is also introduced to make a comparison, and the conclusions indicate BP network performs much better; actually the generalization capability of the former is much better too.