Relabeling for Indoor Localization Using Stationary Beacons in Nursing Care Facilities

Area of Science:

• Computer Science
• Machine Learning
• Ubiquitous Computing

Background:

• Indoor localization is crucial for monitoring staff-to-patient assistance and workload in caregiving settings.
• Limited training data presents a significant challenge for improving the accuracy of machine learning-based indoor localization systems.
• Existing data augmentation techniques may not effectively address class imbalance in Received Signal Strength (RSS) data for indoor localization.

Purpose of the Study:

• To propose and evaluate a novel data augmentation method for machine learning-based indoor localization using Bluetooth Low Energy (BLE) technology.
• To address the challenge of data imbalance in indoor localization datasets by relabeling Received Signal Strength (RSS) data.
• To improve the accuracy of indoor localization for caregiving and nursing staff monitoring.

Main Methods:

• A data augmentation method based on relabeling Received Signal Strength (RSS) data from Bluetooth Low Energy (BLE) beacons.
• Utilizing standard deviation and Kullback-Leibler divergence to identify matching beacons between minority and majority data classes for relabeling.
• Implementing two relabeling variations: full matching and partial matching.
• Evaluating performance using a Random Forest model and weighted F1-score on a real-world dataset from a nursing facility.

Main Results:

• The proposed relabeling method significantly improves the F1-score for minority classes compared to Random Sampling, SMOTE, and ADASYN.
• Full matching demonstrated a 6-8% improvement in the overall weighted F1-score compared to the baseline.
• The augmentation method effectively leverages majority class samples to enhance the training data.

Conclusions:

• The proposed data augmentation technique effectively addresses class imbalance in BLE-based indoor localization datasets.
• Relabeling beacon data offers a promising approach to enhance the accuracy of machine learning models in caregiving environments.
• The method provides a practical solution for improving staff monitoring and workload management through more accurate indoor localization.