Abstract
With the goal of developing procedures for predicting activity/travel patterns of individuals given their socio-demographic characteristics, the authors cluster individuals based on their activity patterns using a two-stage clustering technique to infer activity time windows. The two-stage technique is a combination of affinity propagation and K-means clustering methods. Activity patterns are created by segmenting daily activities into ten-minute intervals, carrying information about activity types, duration, schedule and travel distance. The authors test different combinations of two error measures: sequential alignment and agenda dissimilarity to compute the distance between each pair of patterns. In order to analyze the effectiveness of clustering on inferring activity patterns, the authors further test the prediction accuracy for two population, clustered and un-clustered. The results indicate that updating activity time windows based on the arrival time distribution of the clustered data, has higher accuracy than using those distributions with un-clustered data.