MyWay: Location prediction via mobility profiling

You are here

MyWay, a prediction system which exploits the individual systematic behaviors modeled by mobility profiles to predict human movements. The idea behind MyWay is the possibility for each user to use its own Mobility Profile to predict his momvements, and in case it is not enough access to the collective knowledge. This additional source of information is composed by the mobility profiles of other users sharing their only their models instead of raw trajectory data revealing their detailed movement. In the figure we can see a set of mobility profiles of different users (each one is represented by a different color).

The resulting three predictors are shown in figure 1, for each predictor a different color is used: individual history, the individual profile and the individual predictor (red) are inside the user PMDS, while the collective predictor (blue) is outside and therefore handled by a third party that orchestrates the users’ information as well as the hybrid predictor (green). This third party, usually called coordinator, has the responsibility for the storage and management of the users’ profiles. In the case of the hybrid strategy the coordinator stores all the mobility profiles of the users (which are compact representations of their mobility) and receives the query for the prediction only in the case the individual predictor of a specific user fails. The Hybrid strategy result to be the best solution.

Our tests show the accuracy of the individual predictor exploiting only personal knowledge to predict the future trajectory. The figure 2 reports the performances varying the lookahead, i.e. the distance in time from the moment of the prediction. In the left plot the current movements are cut after 33% of their travel, while on the left are cut after 66%. The performances on the right part are higher, thus having more information the individual predictor works better. In both cases the accuracy drops when looking more than 5 minutes in the future but then stabilizes.

The test in figure 3 compares the various versions of MyWay system, i.e. individual, collective and hybrid against the state of the art competitors for location prediction using an individual and global approach. We can observe how MyWay-hybrid overcomes all the other approaches both in the case of trajectories cut at 33% (left) and 66% (right).

Related publications

Trasarti, R., R. Guidotti, A. Monreale, and F. Giannotti, "MyWay: Location prediction via mobility profiling", Information Systems, vol. 64, pp. 350–367, 03/2017.