On clustering RSS fingerprints for improving scalability of performance prediction of indoor positioning systems

We previously developed an analytical model in [8] to predict the precision and accuracy performance of indoor positioning systems using location fingerprints. A by-product of the model is the ability to eliminate unnecessary fingerprints to reduce the number of fingerprints in a radio map for comparison without loss in performance. This model enables computation of an approximate probability distribution of location selection, by employing a proximity graph to extract neighbor and non-neighbor sets of a given fingerprint. However, employing the model 'as is' in a system with many location fingerprints requires determining a single large proximity graph derived from all location fingerprints which may involve significant computational effort. In this paper, we consider two techniques to divide location fingerprints into smaller clusters. Separate proximity graphs for each cluster are now used to predict performance and eliminate unnecessary location fingerprints. Results show that the computational effort can be reduced, creating a more scalable analytical model, while still predicting and enabling good precision performance.