Table 9.

Summary of the applications of ML for VLC communications.

References	ML Model Architecture	Contributions	Remarks
[137]	Deep RL algorithm	Beamforming control	Significantly increased the secrecy rate, decreased the BER, and outperformed the zero-forcing and other existing algorithms
[138]	GRUs–CNN prediction algorithm	UAV deployment optimization, user allocation, and energy efficiency	Solved the non-convex optimization problem in low complexity and reduced total transmit power by up to 68.9%
[139]	Model-driven DL-nonlinear post-equalizer scheme	Channel estimation and symbol detection	Successfully proved the robustness and generalization ability, compensated for overall channel impairment, and demodulated distorted symbols to bit streams
[140]	ANN-based AE structure	Low-frequency noise effect prediction	Achieved speeds up to 0.325 Gbps faster than another scheme, and robustness to bias, amplitude, and bitrate changes
[141]	LSTM-AE scheme	Sequential data input handling and sequential data output prediction	Significantly reduced the PAPR while maintaining BER