Anomaly Detection for IoT Applications using WW-RF and MLP-DBO Algorithms

The Anomaly detection technique is being utilized aggressively in detection and prevention of various attacks in Internet of Things (IoT) networks.However, it will be difficult to train the dataset containing multiple attack types with huge volumes of records.The associated computational cost and time will be high.In this paper, an anomaly detection technique for IoT applications using WW-RF and MLP-DBO algorithms is proposed.It consists of optimal feature selection and classification techniques.Initially, each kind of attack is given its own dataset by separating it from other attacks.For optimal feature selection depending on their weights, WildWood Random Forest (WW-RF) algorithm is applied.After applying this algorithm, the top K features with the maximum relevance value will be chosen for each attack type.For attack detection and classification, Multilayer Perceptron (MLP) is applied to each file.The metaheuristic optimization algorithm Dung Beetle optimizer (DBO) is applied, to optimise the weights and biases of the MLP.Experimental results show that MLP-DBO attains nearly 96%-100% accuracy for all kind of attacks.It performs significantly better than the existing ML techniques and existing ensemble techniques.