Development of an Improved Model of the Point Location Problem in Indoor Spaces with Fingerprint and Artificial Neural Network Approaches

This study proposes an approach to indoor positioning that combines the RSSI-based fingerprinting method with a dual artificial neural network (ANN) architecture. Unlike conventional approaches that use only RSSI vectors, this approach also incorporates the location information (x, y) of the measurement points into the learning process. The proposed structure consists of two ANNs. Model 1 can predict y′ from x and RSSI inputs. Model 2 can predict x′ from y and RSSI inputs. The outputs of both models are combined. The closest matching point pair represents the estimated position. The proposed approach increases sensitivity to changes in position and improves robustness against multipath effects and environmental noise. Measurements obtained from a simulation environment were prepared and divided into training, validation, and test sets. Multi-layer feedforward artificial neural networks were then trained. The results of the regression and residual analyses showed that both models produced similar results and that the final prediction made using the intersection was reliable. The results show that the suggested bidirectional ANN–fingerprint framework is a scalable, flexible, and data-driven solution that does not require additional hardware. It provides reliable and generalizable position estimation in the simulated indoor scenario considered.