Impact of altering data granularity levels on predictive modelling: a case study of fall risk prediction in older persons