Detecting Reconnaissance Activities in IoT Networks Using UNB CIC-IOT 2023 Dataset
Anshika Sharma, Himanshi Babbar, Amit Kumar Vats · 2024
Cyberattacks are becoming more common at an exponential rate, coinciding with the IoT ecosystem’s stratospheric ascent. Reconnaissance attacks are a crucial part of these dangers because they precede a lot of other negative activities. Because IoT devices and their communication protocols are so diverse and unique, reconnaissance attacks in IoT networks can be hard to detect. This research aims to provide an approach to reconnaissance threat identification that makes use of machine learning (ML). This research makes use of the UNB CIC-IOT 2023 dataset, which offers realistic settings for IoT network traffic. The data set is cleaned up to remove unnecessary information and identify important features before further processing. The next step is to build detection models and test them using various ML approaches including Naive Bayes (NB), Logistics Regression (LR), Adaptive Gradient Boosting (AdaBoost), and Extreme Gradient Boosting (XGBoost). Research shows that XGBoost provides the highest degree of accuracy. From what we can tell, a few of ML algorithms have promising detection capabilities. Different metrics, like as accuracy, recall, F1-score, and precision, have been utilised to assess the predictive usefulness of these models. It has been demonstrated through the data that the strategy that was utilised accurately identifies potential threats to the network and improves the capabilities of Internet of Things devices. Even though, LR, and AdaBoost each had accuracy rates of 71.42%, 86.45%, and 89.21%, respectively, the research showed that XGBoost had the highest accuracy, with a rate of 95.95% on average.