Received Signal Strength Fingerprinting-Based Indoor Location Estimation Employing Machine Learning

Area of Science:

• Computer Science
• Electrical Engineering
• Machine Learning

Background:

• Indoor positioning systems commonly use fingerprinting based on signal strength measurements.
• Traditional methods suffer from reliability issues due to inaccurate power measurements in wireless systems.
• Existing studies on machine learning for radio fingerprinting are difficult to compare directly.

Purpose of the Study:

• To improve the accuracy and reliability of indoor positioning using Bluetooth Low Energy (BLE) fingerprinting.
• To explore the impact of multiple anchors and radio channels on positioning accuracy.
• To evaluate the performance of various machine learning algorithms for BLE-based localization.

Main Methods:

• Utilized multiple anchors and radio channels to extend power level measurements.
• Investigated different alignment approaches for received signal strength (RSS) measurements.
• Implemented and analyzed four supervised machine learning techniques: k-Nearest Neighbors, Support Vector Machines, Random Forest, and Artificial Neural Network.

Main Results:

• The Random Forest algorithm demonstrated superior performance, achieving a classification accuracy exceeding 99%.
• The study analyzed the accuracy-complexity trade-off for candidate algorithms in 1D and 2D environments.
• A comprehensive literature survey identified challenges in comparing existing machine learning applications in radio fingerprinting.

Conclusions:

• Enhancing BLE fingerprinting with multiple anchors and channels, combined with machine learning, significantly improves indoor positioning accuracy.
• Random Forest emerges as a highly promising technique for accurate and reliable BLE-based localization.
• The findings provide valuable insights into algorithm selection based on accuracy and complexity requirements.