Abstract :
Nowadays the significant concern in IoT infrastructure is anomaly and attack detection from IoT devices. Due to the advanced technology, the attack issues are increasing gradually. There are many attacks like Data Type Probing, Denial of Service, Malicious Operation, Malicious Control, Spying, Scan, and Wrong Setup that cause the failure of the IoT-based system. In this paper, several machine learning model performances have been compared to effectively predict the attack and anomaly. The performance of the models is compared with evaluation matrices (Accuracy) and confusion matrix for the final version of the effective model. Most of the recent studies performed experiments on an unbalanced dataset; that is clear that the model will be biased for such a dataset, so we completed the experiments in two forms, unbalanced and balanced data samples. For the unbalanced dataset, we have achieved the highest accuracy of 98.0% with Generalized Linear Model as well as with Random Forest; Unbalanced dataset means most of the chances are that model is biased, so we have also performed the experiments with Random Under Sampling Technique (Balancing Data) and achieved the highest accuracy of 94.3% with Generalized Linear Model. The confusion matrix in this study also supports the performance of the Generalized Linear Model.
Keyword :
IoT Attacks, IoT Anomalies, Random Under Sampling, Machine Learning