Detecting activity type from gps traces using spatial and temporal information

Detecting activity types from GPS traces has been important topic in travel surveys. Compared to inferring transport mode, existing methods are still relatively inaccurate in detecting activity types due to the simplicity of their assumptions and/or lack of background information. To reduce this gap, this paper reports the results of an endeavour to infer activity type by incorporating both spatial information and aggregated temporal information. Three machine learning algorithms, Bayesian belief network, decision tree and random forest, are used to investigate the performance of these approaches in detecting activity types. The test is based on GPS traces and prompted recall data, collected in the Rijnmond region, The Netherlands. Results show that the random forest model has the highest accuracy. The model incorporating spatial and temporal information can predict activity types with an accuracy of 96.8% for the used dataset. These findings are expected to benefit research on the use of GPS technology to collect activity-travel diary data.