Artificial Neural Network-based Estimation of Individual Localization Errors in Fingerprinting

Location information is among the main enablers of context-aware applications and wireless networks. Practical localization services are able to generate location estimates that are generally erroneous. To maximize its usability and benefits, each location estimate should be leveraged jointly with the corresponding estimate of its localization error. Hence, we propose an Artificial Neural Network (ANN)-based method for the estimation of individual localization errors. We do that for fingerprinting, one of the most prominent localization solutions for GPS-constrained environments. First, we provide insights on how to optimally hyperparameterize the proposed method. We do that by exploring its hyperparameter’ space in order to find its close-to-optimal hyperparameterization for different environments and fingerprinting technologies. We believe the provided insights can serve to reduce the overhead of deploying the method in new environments. Second, we demonstrate that the method, when hyperparameterized according to the provided insights, substantially outperforms the current state-of-the-art. The improvement is more than 25% in the best case scenario.