Measuring the Gap Between the Maximum Predictability and Prediction Accuracy of Human Mobility

It has been claimed that human mobility is highly predictable and an upper bound of 93% predictability is achievable. However, there is a significant gap between the upper bound of predictability and the actual prediction accuracy in many data sets. This paper points out that this gap is caused by the difference between the user's actual distribution and the hypothesis in the derivation through the analysis on the upper bound of predictability. Then two statistics of the target user's mobility traces are proposed to measure this gap, whose effectiveness is validated by simulated traces and real-world data sets using five prevailing prediction models. The proposed MLP statistics can help with assessing the data quality and designing prediction algorithms. Our work makes the predictability upper bound become a more effective measure and extends the understanding of predictability research in human mobility prediction.