New Machine Learning Hybrid Models to Lower Position Errors for Bluetooth-Based Indoor Localizations

In recent times, there has been a significant development of critical IoT-based applications in the field of indoor localization (e.g., locating an asset or person). The utilization of Machine Learning (ML) algorithms to enhance accuracy in the presence of interference has generated considerable interest among researchers in this domain. This research paper introduces two hybrid models, namely the Asymmetric and Symmetric Line-Shifting hybrid models, which integrate novel line shifting algorithms with ML classifiers and regressors. The primary objective of these models is to reduce positioning errors. By shifting learned and predicted Reference Points using ML algorithms, the models calculate centroids that closely align with the targets' locations. Through real-world analysis, it was demonstrated that the new models achieved lower Root Mean Square Error (RMSE) and improved R-squared values compared to those obtained using the Constant Line Shifting algorithm (e.g., 52% and 29%, respectively). Moreover, both models outperformed the version without Reference Point shifting in terms of RMSE and R-squared values (e.g., 76% and 273%, respectively). Consequently, the proposed hybrid models are deemed accurate and reliable for many indoor localization applications.