Machine Learning in Visible Light Communication System: A Survey

Table 3

ML algorithms for position estimation.

Method	Ref.	Outcomes
-NN & -NN	[14–17]	(1) The dataset must be modified with the change in transmission medium; (2) The models are trained with RSS-based datasets; (3) These models outperform conventional methods based on received signal strength; (4) [15] Has the best average location accuracy of 4.2 cm; (5) Low to moderate efficiency
ANN	[23, 41, 46, 47]	(1) A large amount of sampling data is required; (2) Highly reliable after plenty of training process; (3) Sensors or RSS-based datasets train models; (4) With the LoS component, [41] has the best positioning accuracy of 2.7 cm; (5) Useful for real-time location monitoring; (6) Highly efficient
-Means	[42, 43]	(1) Models are trained with RSS-based datasets; (2) Moderate reliability; (3) Less accurate than ANN and -NN; (4) Efficiency moderate to high
Comparison of SVM, RF, -NN, and DT	[17]	(1) Highest location accuracy of 8.6 cm with SVM; (2) Shortest mean calculation time of 5.6 cm with -NN
Multiple classifier	[44, 45]	(1) Take advantage of strengths of single classifier; (2) Precise and reliable; (3) High calculation complexity