Producción CyT
Prediction of the next user location using personal tracking data
Articulo
Autoría:
Vallejos, Sebastian ; Berdun, Luis S. ; Monteserin, Ariel J. ; Godoy, Daniela L. ; Armentano, Marcelo G. ; SCHIAFFINO, SILVIA NOEMIFecha:
2025Editorial y Lugar de Edición:
PERGAMON-ELSEVIER SCIENCE LTDRevista:
EXPERT SYSTEMS WITH APPLICATIONS PERGAMON-ELSEVIER SCIENCE LTDResumen *
The use of mobile devices for the study of urban mobility has attracted the attention of the scientific community in the last years. In this work, we address the problem of predicting the user’s next destination given the user’s mobility profile, i.e. her visit history and the set of places she visited during that time. To predict the next visit, our approach relies on a set of predictors, each of which analyzes different aspects of the user’s daily mobility. The output of each predictor consists of a list of places along with the probability of each place being the next destination. The approach combines the output of each predictor into a single one using weights based on the cumulative precision of its previous predictions. Once the user arrives at her destination the proposed approach adapts the set of predictors for future estimations. The predictors in this work were designed so that their training and use have a low computational cost, adapting to the available resources on mobile devices. The experiments carried out over two datasets (Nokia MDC dataset with 52 users and 129,097 visits to 12,869 different places, and a private dataset with 76 users with more than 30 registered days and 47,933 stays) yielded promising results, enhancing performance for 75.8% of users with an average accuracy of 49.2% and F1 of 42.6% and an average use of only 48 Megabytes of RAM. Información suministrada por el agente en SIGEVAPalabras Clave
MOBILITY PROFILESPROBABILITYPREDICTIONNEXT VISIT