Performance analysis of fingerprinting indoor positioning methods with BLE

• Published in: Expert systems with applications 2022-09, Vol.202, p.117095, Article 117095
• Main Authors: Aranda, Fernando J., Parralejo, Felipe, Álvarez, Fernando J., Paredes, José A.
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; BLE; Bluetooth; Computing time; Fingerprinting; Indoor positioning; Multilayer Perceptron (MLP); Optimization; Parameters; Smartphones; Support vector machines; Support Vector Machines (SVM); Training; Weighted k-Nearest Neighbours (Wk-NN)
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2022.117095

Indoor positioning with smartphone-compatible technologies has fostered much research attention in recent years. In this context, Bluetooth Low Energy (BLE) reveals a good performance for this type of task. It offers more flexibility and better achievements when compared with similar systems based on IEEE 802.11 Wireless LAN (Wi-Fi) technology, especially for fingerprinting-based positioning systems. The literature on these systems is rich and growing; however, not all its possible algorithms have been tested and compared under similar conditions for this emergence technology. This work presents a thorough analysis of the state of the art on Wi-Fi and Bluetooth Low Energy (BLE) algorithms used for fingerprinting systems. Based on this study, a novel scheme for fingerprinting methods classification is proposed. Then, a performance comparison between the Bluetooth Low Energy (BLE) databases is carried out, assessing training time, parameter optimization, computational time, and positioning accuracy. For the sake of completeness, a new database is provided and compared with the others to analyze how the environment can affect the accuracy of each method. The results show that those techniques based on the Weighted k-Nearest Neighbours (Wk-NN) algorithm perform better on average for large scale deployments; besides, they do not require any previous training and consume less time to optimize their parameters. On the other hand, Support Vector Machines (SVM) provides the best accuracy with less computational and training time in small environments. •Wk-NN algorithm shows the best overall performance.•SVM require less training time for optimal accuracy.•Low performance of probabilistic-based algorithms.•Wk-NN is the preferable option for large mixed environments.•SVM is the preferable option for small indoor areas.