Assessing fingerprinting and machine learning approaches for wireless indoor localization

Abstract

This paper presents a comparative analysis of fingerprinting and machine learning techniques for bluetooth low energy (BLE)-based localization. Two fingerprinting algorithms, namely fingerprint feature extraction (FPFE) and Bayesian estimation (BE), along with various machine learning approaches including support vector regression (SVR), ensemble learning, and instance-based learning, are investigated. The selection of techniques depends on the availability of training data or the fingerprint database, explored in both ideal scenario and real-world scenario. In ideal scenario where the system administrator can collect fingerprint data through users’ devices, FPFE emerges as the preferred algorithm, achieving superior performance with a mean error of 0.50 m. In the context of real-world scenario, where data collection from multiple devices is limited, the system administrator may gather fingerprint data for localization using one or a few specific devices. Our experiments reveal that when there is a scarcity of fingerprint data, BE and SVR exhibit acceptable performance, reaching a mean error of 1.785 m and 1.965 m, respectively.