Feature Entropy Estimation (FEE) for Malicious IoT Traffic and Detection Using Machine Learning

Table 1

Overview of reviewed IDS for IoT security.

Ref	Techniques	Attack types	Dataset	Drawbacks
[10]	Machine learning	DoS, man-in-the-middle attack, spoofing, reply attack, etc.	—	Does not support real-time detection
[12]	Online sequential extreme learning machine	DoS, R2L, probe, U2R	NSL-KDD	It cannot analyze all kinds of attacks evolving in the highly dynamic IoT environment
[13]	Autoencoders	DoS, R2L, probe, U2R	NSL-KDD, KDD99, real time	Not suitable for multiclass attack scenarios
[14]	Social leopard algorithm	Ransomware attacks	UNSW-NB15	Only applicable on ransomware attacks
[15]	Support vector machine	DoS attacks	CICIDS2017	Not suitable for changing traffic flow.
[16]	Machine learning	Port scanning, HTTP and SSH brute force, and SYN flood attacks	Real IoT testbed	Operable on limited data rate of incoming packets
[17]	Random forest	DoS, R2L, probe, U2R	KDD99	Does not support real-time detection
[18]	Deep feedforward neural network	DoS, R2L, probe, U2R	NSL-KDD	Does not support real-time detection
[19]	Convolutional neural network	Flooding DDoS attack	Real IoT testbed	Training error shows steep convergence curve
[20]	Machine learning	Botnets attacks	Bot-IoT dataset	Not suitable for multiclass attack scenarios