Winner Team of the Big Data Challenge at BigSurv20 conference

Our team won the big data challenge of the BigSurv20 conference. Specifically, we worked on the “predicting travel motives” challenge, where the goal is to predict travel motives (e.g. work, shopping, visit friends) from raw GPS data. We took the approach of enriching the GPS coordinate data with external information such as point-of-interest data (e.g. the number of or distance to nearby shops, schools etc.), real-time weather and timing (e.g. time of the day and day of the week), and trained multiple machine learning models that achieved reasonable prediction accuracies (> .80). In addition, we developed a shiny-based app to crowdsource the labelling of travel motives (especially for unannotated data sets).