An Evaluation of Machine Learning Classifiers for Prediction of Attacks to Secure Green IoT Infrastructure

Internet of things is an emerging technology that allows many devices to be connected in an unparalleled way. Despite having many beneficial applications, IoT technology presents significant emission risks due to the large number of devices used in the applications. Therefore, to gain maximum benefit from IoT, we must step towards green IT. On the other hand, cloud computing has been successfully used to provide limitless computational storage and other resources for a variety of IoT devices across the internet. Unfortunately, security concerns in cloud computing for IoT are still a concern. Motivated by the goal of creating a better atmosphere for IoT and ensuring its resilience to risks and attacks, this report reveals ways to decrease the impact of energyuse by IoT on the environment. Additionally, it addresses research concerns for IoT security and reflects on how to protect green IoT networks through the use of an effective machine learning intrusion detection technology to deter attacks on IoT platforms. To do that, we first evaluated some existing ML classifiers such Artificial Neural Network (ANN), Support Vector Machine (SVM), Gaussian Naïve Bayes (NB), Decision Tree (DT) and Random Forest (RF) with the old KDD'99 datasets. The accuracy was extremelyhigh for all classifiers except Gaussian NB whose accuracy was < 90%.The SVM is the highest at 99.24% accuracy with a loss of 4.68% in the last epoch of training. However, using a more recent dataset (ISCX1DS2012) on these same ML classifiers, we observed some discrepancies, all the classifiers dropped in their predictive accuracy even after altering the hyper-parameters.The ANN was at its lowest accuracy at 85.92% and the SVM which was relatively accurate dropped to 90.02%. NB algorithm produced approximately 67.9% accuracy which made it less accurate for both datasets. Based on these findings, we proceeded to propose an efficient model with enough hidden layers and nodes to increase the detection accuracy and to outperform the existing ML classifiers when evaluated with a more recent dataset