Light Gradient Boosting Machine-Based Link Quality Prediction for Wireless Sensor Networks

Table 1

Summary of related work on link quality prediction.

Type	Name	Input	Output	Strength	Weakness
LQP based on link characteristics	Two-stage model [11]	LQI	Classify link as reliable or weak	Quickly determining whether the link can be used.	The remaining links need more testing to be classified.
	RADIUS [12]	RSSI	Classify link as good or weak	It can adapt to dynamic environment changes.	The model accuracy is not high.
LQP based on statistics	FLS [13]	LQI, RSSI, and ER	Classify link as very low, low, medium, or high	It defines a general guideline and can be applied on other routing protocols.	The model accuracy is not high.
LQP based on machine learning	XGBoost-LQP [14]	RSSI, LQI, and SNR	Classify link as bad, medium, or good	Data imbalance is addressed.	The model has high overhead.
	RVFL-LQP [15]	SNR	Probability-guaranteed interval boundary of SNR	The dynamic stochastic features of link quality are described.	The results of the model need to further determine whether the link is available.
	RNN-LQP [16]	LQI	LQI	The temporal correlations of physical layer parameter series are considered.	The model has high time complexity.