Deep embedding approach to classify purpose of trips between cities from GPS data

Abstract

I present a computational framework to identify purpose of trips between cities from GPS traces using a deep embedding approach. I extracted statistical features that captures trips characteristics that includes: temporal features, spatial features and Points of Interests (POI) features. I deployed a deep learning model to extract representative features in a lower dimensional space, which I then feed to a classic clustering algorithm to uncover purpose of trips. I detected six main purposes from trips coming from five different metropolitan areas in the United States to New York city. The trips' purposes detected are: work, which is the most dominating in size, entertainment, shopping, academic, and travelling. I interpret and discuss each cluster in terms of its features. I also compare cities from which trips originated by the distribution of their trips purposes.