Utilizing Machine Learning for Indoor Localization with Multiple Wi-Fi Assistance

This paper evaluates the efficacy of diverse indoor positioning methods, highlighting the superior performance of the Backpropagation (BP) algorithm. Through extensive experimentation, we compared the Mean Square Error (MSE) and Cumulative Distribution Function (CDF) of positioning errors across different algorithms, including Fingerprinting and Triangulation. The BP algorithm exhibited significantly higher accuracy and stability, with a CDF of approximately 0.92 at a 1-meter MSE, surpassing Fingerprinting and Triangulation, which achieved CDFs of 0.77 and 0.06, respectively. Moreover, we examined the impact of varying
$K$
values in the Fingerprinting method based on
$K$
-Nearest Neighbors (KNN) and different configurations of hidden layers and neurons in the BP algorithm. Optimal performance was observed with
$K=2$
for Fingerprinting, yielding an MSE of 1.5 meters, and with 22 hidden layers and 47 neurons for the BP algorithm, resulting in an MSE of 0.84 meters. These findings underscore the significance of parameter optimization for enhancing positioning accuracy and suggest potential avenues for future research aimed at further enhancing indoor positioning systems.