An Effective Machine Learning Approach for loT Intrusion Detection System based on SMOTE

Intrusion detection system to secure loT is a sophisticated tool to detect possible intrusions in the network and ensures confidentiality, integrity, and availability. loT is a precious domain that improves the standard of life, which cannot be accomplished in the existing conventional paradigm. The intrusion detection system is effective in identifying whether the attack is normal or not. Thus, classification algorithms can be applied for prediction. The Machine Learning and Deep Learning concepts of AI technology which contribute more to Data Science have produced remarkable developments in loT applications. In this paper, Machine Learning (ML) algorithms are used to secure loT devices using intrusion detection systems while working on loTID20 dataset. This dataset is highly imbalanced and contains different types of attacks and sub-attacks. The effect of the oversampling technique, Synthetic Minority Oversampling Technique (S MOTE) to balance the dataset significantly, has influenced the result. loT ID20 is a supervised dataset and different classification algorithms are used to measure the performance metrics namely, Accuracy, Recall, Precision, and F-score. The Binary and Multi classifications are done on the dataset using ML techniques. It is found that the accuracy obtained using the ML classifiers such as K-N earest Neighbor, Decision tree and Random Forest techniques is above 90%, showing that the mitigation of attacks that occur on an loT network is effective.