This work aims at reducing the distance estimation error when using the Bluetooth signal intensity. One of the main objectives is to improve indoor positioning by inferring distances more precisely. The methodology includes the generation of a large signal-distance sample dataset, which is processed and analyzed with several machine learning algorithms, previously applying a series of mobile average and smoothing signal filters. The best combination of filter and algorithms was obtained by first applying a moving average with a window of 115 samples, followed by an M5P algorithm integrated in a Bagging meta-algorithm. Finally, the mean average error, calculated by cross-validation over distances of up to 10 m, was 168 cm, which substantially improves previous experiments focused on similar techniques.
Paper
Full text
Increasing Accuracy in Positioning by RSSI: an Analysis with Machine Learning Algorithms
Semantic Scholar · Computer Science · 2019
Abstract
This work aims at reducing the distance estimation error when using the Bluetooth signal intensity. One of the main objectives is to improve indoor positioning by inferring distances more precisely. The methodology includes the generation of a large signal-distance sample dataset, which is processed and analyzed with several machine learning algorithms, previously applying a series of mobile average and smoothing signal filters. The best combination of filter and algorithms was obtained by first applying a moving average with a window of 115 samples, followed by an M5P algorithm integrated in a Bagging meta-algorithm. Finally, the mean average error, calculated by cross-validation over distances of up to 10 m, was 168 cm, which substantially improves previous experiments focused on similar techniques.