Comparative analysis of fingerprint-based models for indoor visible light positioning: superior performance of ANN over KNN and DNN

This study compares the performance of three fingerprint-based models—K-Nearest Neighbors (KNN), Artificial Neural Network (ANN), and Deep Neural Network (DNN)—for indoor Visible Light Positioning (VLP) systems. The simulation aims to determine the most accurate and reliable method for precise indoor positioning applications. The results indicate that the Fingerprint-ANN model outperforms both the Fingerprint-KNN and Fingerprint-DNN models in terms of positioning accuracy. With a maximum error of 0.0136 m, a minimum error of 1.5449 × 10⁻⁵ m, and an average error of 0.0016 m, the Fingerprint-ANN model achieved significantly lower errors compared to the Fingerprint-KNN (maximum error: 0.1803 m, minimum error: 0 m, average error: 0.0548 m) and Fingerprint-DNN (maximum error: 1.1106 m, minimum error: 1.6236 × 10⁻⁵ m, average error: 0.6067 m) models. The superior performance of the Fingerprint-ANN model can be attributed to its ability to capture complex non-linear relationships within the signal strength data, leading to higher accuracy and reliability. Furthermore, the resemblance of the positioning error mesh plot of the Fingerprint-DNN model to that of the trilateration method suggests that while DNNs can model complex relationships similar to traditional methods, they might require more data or tuning to reduce error further. Overall, these findings highlight the potential of the Fingerprint-ANN model for achieving enhanced precision in indoor VLP systems, making it the preferred method for applications requiring high accuracy and reliability.