Temporal Data Simulation based on a real data set for fall prevention
Résumé
The motivation of this article is the need for a large temporal
data set, which allows to reason about the changes regarding person’s features over a long period of time. Faced with
the difficulty of finding such a data set, we propose an algorithm to simulate such a data set, based on real static data provided by the service of fall prevention of Lille’s hospital. We
select five persistent variables, meaning that their value may
change at most once, toward positive value for positive persistent variables. The algorithm is based on assumptions regarding the temporal evolution of each contextualized variable, as
defined by a Bayesian network learned on the real static data
set. The temporal data set simulated thanks to the proposed
algorithm is evaluated by the comparison of the temporal distribution of each contextualized variable with the functions
obtained by linear interpolation from the real data set.