Understanding the impact of data sparsity and duration for location prediction applications

Abstract

As mobile devices capable of sensing location have become pervasive, the collection and transmission of location data has become commonplace, enabling the creation of models of behaviour that support location prediction. With such devices often heavily resourceconstrained, the nature of data used in location prediction must be understood in order to optimise storage and processing requirements. This paper specifically explores data sparsity and collection duration. The results presented provide insight which suggest: (i) a relationship of diminishing returns in predictive accuracy when collecting user location data at increased rates over a fixed period, and (ii) the duration over which a fixed size sample of location data is collected has a greater impact on predicative accuracy than data sparsity.