Features selection and prediction for IoT attacks

Cyber-attacks and anomaly detection are growing concerns in the Internet of Things (IoT). With fast-growing deployment and opportunities, an increasing number of attacks put IoT devices under the threat of continuous exploitation and danger. Malicious operation, denial of service, MITM, and scan are major types of attacks that can cause IoT devices to fail. We study how a variety of machine-learning algorithms, such as decision tree, random forest, and gradient-boosting machine (GBM) analyze and predict network attacks on IoT devices. By comparing performance indicators for various algorithms through different model evaluations, we conclude that a decision-tree algorithm is generally the most accurate compared with random forest and gradient-boosting machine, but the random forest algorithm has better AUC scores as it combines the results of multiple individual decision trees. Gradient-boosting machine performs well, but from an accuracy and time aspect, it may not be the best option.