Table 9.

Qualitative comparison of smart transportation-related works.

	Use Case	Ref	Contribution	AI Role (At the Edge)	AI Algorithm	Dataset	AI Placement	Employed Technology	Platform	Metrics	Benefits AI-Edge	Drawbacks
Smart transportation	SPM	[111]	Real-time prediction Bike charging at each stationReduce load to cloud	Prediction	RT SOM	Kaggle competition, London shared bike data	MEC	Lightweight model (ML)	Not mentioned	RMSE RMSLE	High accuracy Generalization	Multivariate data not supported security
		[112]	Real-time parking occupancy surveillance Reduce load to cloud	Classification	Mobile-net SSD, BG, SORT	MIO-TCD	Edge device Raspberry Pi 3B,	Transfer learning	TensorFlow Lite	Accuracy	Flexibility Reliability Online and high accuracy	Accuracy needs to be enhanced (=95),security
		[113]	Privacy preserving Parking space estimation	Prediction, decision making	LSTM DRL Game theory	Birmingham parking dataset	Fog nodes	Federated learning	Not mentioned	MSE	Computation offloading in nonstatic environment, improve security, flexibility, high accuracy	Less convergence speed
	T.M.P	[115]	Timely citywide traffic prediction, context data management	Data aggregation	CNN, LTSM	Beijing taxicabs data NYC bike data	Fog nodes	Transfer learning	IFogSim	Complexity, training time, prediction time, accuracy	Reduce network congestion,increase energy efficiency, less training/prediction times	Cloud inference, non-real-time prediction
		[114]	Forecast the overall traffic, adjust the redirected flow	Prediction	DBN-SVR	Caltrans PeMS	Fog nodes	/	MATLAB	Scalability, processing time, accuracy	Scalability, security	Accuracy needs to be enhanced
		[116]	Privacy preservation Traffic flow prediction	Prediction	GRU, k-means	PeMS database	Edge nodes	Federated learning	Not mentioned	MAE, MSE, RMSE, MAPE	Low communication overheadStatistical heterogeneity solved, high accuracy	Spatiotemporal correlation not solved
		[117]	Timely traffic flow prediction	Prediction	SVM PSO	Guiyang City dataset	Fog nodes	Lightweight ML	Matlab 2014a	MSE	Low time overhead, faster processing, adaptability, good prediction	Model complexity high
		[118]	Spatial traffic flow prediction	Prediction	GCNs	TaxiBJ TaxiNYC dataset	Edge nodes	Federated learning	Not mentioned	RMSE, MSE, MAPE	High accuracy	Less scalability
	ITM	[88]	Driver distraction identification	prediction	VGG1-CNN -k-means	Kaggle’s state farm, distracted driver challenge	Edge deviceRaspberry Pi	Transfer learning	KERAS	Accuracy, precision, recall, F1-score	High accuracy	Securityless scalability
		[119]	Driving behavior evaluation	Prediction	CNN-LSTM	ToN UCI knowledge discovery, archive database	Fog nodes	Transfer learning	TensorFlow	Accuracy-loss curves	High accuracy, generalization	Less scalability, security
		[120]	Real-time fault diagnosis	Prediction	SAES-DNN, knowledge graphs	ToN UCI knowledge, discovery archive database	Edge deviceNVIDIA Jetson TX2	Transfer learning	Python	Loss rate accuracy	High accuracy	Model complexity, accuracy degraded for largedataset