Table 7.

Qualitative comparison of smart industry 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 Industry	FI	[69]	Financial data analysis	Prediction	SVM	(Credit card fraud, credit card risk, Customer Churn, Insurance Claim) dataset	Edge devices, cloud	Low-cost model task offloading	Simulator (Not mentioned )	Task assignment over delay power consumption precision recall F1-score	High accuracy	Communication overhead
		[70]	Early-warning of financial risks	Prediction	BPNN	Real-world dataset	MEC	Quantization HARDWARE-CPU	Matlab	Accuracy, hit rate	Less response time	Accuracy needs improvement
	C.I	[71]	Locality-based product demand prediction and decision making	Feature selection, classification, decision-making	RL, PCA, K-means	Kaggle open data	Edge device (GPU NVIDIA-SMI)	Low-cost model	Scikit-learn Python	Clustering score maximum/average cumulative reward execution time	Outperform others existing methods	Stability not tested
	MMM	[72]	Machine malfunction monitoring	RF SVM Adab LR MlP	(MIMII dataset	Fog (controller unit (ICU)/Microdata center)	Hardware accelerator	Lightweight model	Not mentioned	Time complexity, accuracy, precision, FScore	Response time reduction	–
		[73]	Abnormal events detection during assembly line production	Outlier detection prediction	RF, DBSCAN	Real-world dataset	Edge devices (Raspberry Pi)	Low-cost model	MongoDB Python	Accuracy recall F1-score precision	High accuracy	Dynamic of IoT data not addressed
		[74]	Fault detection in a hydraulic system	Data reduction classification	LSTM, AE, GA	Real-world dataset	Edge server	Transfer learning	TensorFlow	Complexity DL accuracy detection time, data reduction	Reduction of load to cloud Low detection time Robust to noisy data	Communication overhead
Smart Industry	MM	[75]	Faults of machine detection	Classification	LSTM	Real-world dataset	Edge device (Raspberry Pi)	Lightweight model	Keras Python	Accuracy	Low-cost model Short fault detection	Memory usage overhead
		[76]	Fast manufacture inspection	Feature extraction classification	CNN	Real-world dataset	Fog gateway	Early exit-DNN splitting	Not mentioned	ROC curve running efficiency	High accuracy	High communication cost
	PQMP	[77]	Fast prediction of assembly quality	Feature selection, prediction	RF Adaboost	Real-world dataset	Edge server (PC)	Transfer learning	Python	Accuracy	Efficacy flexibility complexity reduction	Online learning not improved
		[78]	Fast tool wear monitoring and prediction	Feature extraction classification	CNN LSTM BiLSTM	Real-world dataset	Edge server (PC)	Transfer learning	Python TensorFlow	Response time, network bandwidth, data transmission RMSE MAPE	High monitoring accuracy, low-cost model, low response latency	Accuracy loss
		[79]	Scheduling tasks production for smart production line	Task scheduling, resource allocation	PSO, ACO	Not mentioned	Fog gateway	-	Matlab	Completion time, energy consumption, reliability	Solves the problem of limited computing resources, high energy consumption, real-time/efficient processing	Does not consider heterogeneity of IoT devices.