Elephant Flows Detection Using Deep Neural Network, Convolutional Neural Network, Long Short-Term Memory, and Autoencoder

Table 1

Summary of previous works.

Paper	Networking problem	Approach	Method/classifier technique	Application type	Dataset used	Experimental result	SDN	ML/DL	Elephant	Data center
Aceto et al. [30]	Encrypted TC	Deep learning	SAE, CNN, and LSTM	HTTP traffic	300 k Mobile datasets activity	SAE outperforms in TC	⁰	•	⁰	⁰
Nascita et al. [35]		Deep learning	1D-CNN	WeChat	MIRAGE 2019 dataset	Global model interpretation r	⁰	•	⁰	⁰
Ali[37]	Intelligence TC	Deep learning	Deep neural network	HTTP, mail, and multimedia	Moore dataset	Better accuracy, precision, recall, and FScore results	⁰	•	⁰	⁰
Dong and Xia [39]	Deep learning application	Deep learning	SC-CNN	Image segmentation	MNIST dataset	Deep learning create more powerful optimization methods	⁰	•	⁰	⁰
Dong et al. [38]	Abnormal traffic detection	Deep learning	Kmean, AE, and reinforcement	Abnormal traffic	NSL-KDD and AWID datasets	Achieved good result in time complexity	⁰	•	⁰	⁰
Dong et al. [36]	Tc	Deep learning	CNN and GAN	FTP, Gmail, and Skype	USTC-TFC2016 dataset	Yields better application traffic classification and detection result	⁰	•	⁰	•
Bovenzi et al. [29]	Model parallelism TC	Machine learning	RF, DT, and Bayes	IoT and fog platform	Anon17 NIMS dataset	Reducing training time	⁰	•	⁰	•
Chang [40]	Encrypted TC	Machine learning	RNN-AE	Github, Gmail, and Icloud	Real-world dataset 18 applications	Achieves 99.14% performance	⁰	•	⁰	⁰
Ibrahim [49]	Online game	Machine learning-mixed	Fixed Java code	http, FTP, and Skype	LOL game dataset	Produces 91% accuracy	⁰	•	⁰	⁰
Dong [20]	Multiclass TC	Machine learning	SVM	http, imap, and dns	MOORE and NOC datasets	Improve classification accuracy	⁰	•	⁰	⁰
Shi et al. [50]	Machine learning Application	Machine learning	Netflow extended machine learning	FCBF algorithm	Machine learning Application optimization	Algorithm called FCBF yields better performance	⁰	•	⁰	⁰
Dong and Li [42]	Traffic identification	Machine learning	Neural network and Naïve Bayes	TCP and UDP flows	MOORE and NOCSET	Achieves 95% identification accuracy	⁰	•	⁰	⁰
Shi [13]	Online encrypted	Machine learning	Naïve Bayes	Online Skype	Skype-SET	Reduces false positives and false negatives	⁰	•	⁰	⁰
Dong et al. [16]	Traffic identification	Algorithm based on architecture	High identification accuracy	Routing application	NOC_SET, CAIDA, and LBNL_SET	High accuracy	⁰	•	⁰	⁰
Oliveira [2]	QoS	Architecture	Fixed Python program	Distributed applications	Small text and video dataset	Low overhead	•	⁰	•	•
Hamdan et al. [11]	Load balancing	Architecture	Fixed Python program	d/t applications	Sketch-based filter elephants	Good running time and performance	•	⁰	•	•
This paper	Elephant flow detector	Deep learning	DNN, CNN, LSTM, and AE	Real-time apps	NIMS and SDN datasets	98.78% accuracy	•	•	•	•