Imputing Trip Purpose Based on GPS Travel Survey Data and Machine Learning Methods

摘要: In the recent decades, increasing number of travel researchers show interest in behavior research based on GPS/GIS technology. The challenge successfully utilizing GPS-based data is efficient post-processing method that could generate essential components as accurately possible researches such time, trip purpose, mode, and length. This paper concentrates part GPS post-processing: purpose derivation, explores feasibility automating detection employing machine learning with geospatial location data, land use in-practice survey conducted by University Minnesota. Furthermore, it evaluates impacts different coding methods polygon-level, geo-coded home/work locations Point Interest (POI) combined including decision tree, support vector metalearner. A heterogeneous sample 2238 records decoded 7 purposes employed. Results under all methods, cluster-based exceeded closest POI method, while amongst three metalearner has best performance to classify purpose. Based set using highest classification accuracy 80.5817% can be achieved.