Work

BLE-Based Indoor Localization System for Parking Area using Fingerprinting Machine Learning

Year: 2026
Type: article
Abstract: Indoor localization systems for basement parking areas face significant challenges due to satellite signal limitations and complex signal propagation in enclosed environments. This problem can be addressed through RSSI fingerprinting approach based on Bluetooth Low Energy with machine learning classification. This research implements a localization system based on Light Gradient Boosting Machine optimized through the Optuna framework to classify 25 discrete locations in a five-by-five grid layout within a 10-by-10-meter parking area. Implementation uses one ESP32-S3-based mobile beacon, four anchor nodes with five dBi external antennas positioned at area corners, and Raspberry Pi 4 as computational platform. RSSI data was collected systematically with fifty samples per cell, resulting in a dataset of 1250 samples extracted to generate thirty-six statistical features including mean, median, standard deviation, and quartiles. Hyperparameter optimization was conducted through one hundred Optuna trials with objective function maximizing macro F1-score via five-fold stratified cross-validation. Evaluation on independent test set demonstrates 99.6 percent accuracy with only one misclassification from 250 samples, and inference time of 0.004 milliseconds per sample on Raspberry Pi 4. These findings indicate that the combination of LightGBM with Optuna optimization and statistical feature extraction is effective for medium-scale parking localization but requires further validation under varying environmental conditions.
Language: English
Open Access status: hybrid
OpenAlex ID: W7236784588
FWCI:
Cites:
Cited by:
Related to: 10