RELoc: An Enhanced 3D WiFi Fingerprinting Indoor Localization Algorithm with RFECV Feature Selection

Area of Science:

• Computer Science
• Electrical Engineering
• Geomatics Engineering

Background:

• WiFi fingerprinting is crucial for indoor localization but often limited to 2D.
• Existing 2D methods struggle with vertical ambiguity and spatial modeling in multi-floor environments, reducing reliability.
• Accurate indoor localization is essential for numerous real-world applications.

Purpose of the Study:

• To propose RELoc, a novel 3D indoor localization framework addressing the limitations of 2D approaches.
• To enhance WiFi fingerprinting-based localization accuracy and reliability, particularly in complex multi-floor settings.
• To integrate advanced machine learning techniques for optimal Access Point selection and precise coordinate regression.

Main Methods:

• Utilizes Recursive Feature Elimination with Cross-Validation (RFECV) for efficient Access Point (AP) selection.
• Employs Extremely Randomized Trees (ERT) for accurate 2D and 3D coordinate regression.
• Optimizes ERT hyperparameters via Bayesian optimization with Optuna's Tree-structured Parzen Estimator (TPE).

Main Results:

• RELoc achieves superior performance in both 2D and 3D indoor localization on SODIndoorLoc and UTSIndoorLoc datasets.
• Demonstrates Mean Absolute Errors (MAEs) of 1.84 m (2D, SODIndoorLoc) and 4.39 m (2D, UTSIndoorLoc).
• Achieves significant improvements (33.15% and 26.88%) when incorporating floor information and outperforms state-of-the-art methods like GNN, DNN, and ET.

Conclusions:

• The proposed RELoc framework significantly enhances indoor localization accuracy and robustness, especially in multi-floor environments.
• 3D spatial modeling is critical for achieving spatially discriminative and reliable indoor localization.
• RELoc offers a promising solution for precise indoor positioning using WiFi fingerprinting and advanced AI.