Explaining human mobility predictions through a pattern matching algorithm

Abstract Understanding what impacts the predictability of human movement is a key element for further improvement mobility prediction models. Up to this day, such analyses have been conducted using upper bound mobility. However, later works indicated discrepancies between and accuracy actual predictions suggesting that estimation not accurate. In work, we confirm these and, instead measure, focus on explaining predictions. We show dependent similarity transitions observed in training test sets derived from data. propose evaluate five pattern matching based-measures, which allow us quickly estimate potential As result, find our metrics can explain up 90% its variability. also measures were proved variability fail accuracy. This suggests measure should be compared. Our used assess how predictable data will algorithms. share developed as part HuMobi, open-source Python library.