PCA, Random-Forest and Pearson Correlation for Dimensionality Reduction in IoT IDS

The exposure of Internet of Things (IoT) networks led to an exponential increase in traffic size. Thus, the traffic data meet the 5V big data model, which results in the rise of many cyber threats and the appearance of new security challenges. Intrusion Detection Systems (IDS) need to apply dimensionality reduction to handle the enormous data size and to speed-up cyber threats detection. This research analyzes the impact of Principal Component Analysis (PCA), Random-Forest (RF), and filter-based dimensionality reduction methods on several detection models in four different datasets. Additionally, it studies the impact of some parameters on dimensionality reduction methods’ performance. Results showed that dimensionality reduction methods were able to reduce datasets’ size without compromising the detection models’ efficiency. Moreover, results showed that PCA was the best method in reducing datasets’ size over all the datasets. At the same time, RF showed a better distinguishing ability with the minimum amount of information.