Analysis of the Impact of Dimensionality Reduction on the Accuracy and Performance of Intrusion Detection Systems in IoT Environments

Abstract

The Internet of Things (IoT) has transformed technology by facilitating seamless com- munication and data exchange among interconnected devices. However, this in- creased connectivity poses security challenges, necessitating intrusion detection sys- tems (IDS) to protect IoT environments. This study examines the influence of dimen- sionality reduction methods on IDS accuracy and performance in IoT. We analyze various dimensionality reduction techniques and their impact on IoT intrusion detec- tion systems. Four machine learning models (linear regression, decision tree, SVM, MLP) are implemented with principal component analysis (PCA) as the chosen re- duction method. The IoTID20 dataset is used for training and testing. Comparative evaluations with existing algorithms measure accuracy, F1-score, fit time, and score time. Results reveal that PCA significantly reduces training time without significant accuracy loss. This research offers insights into the impact of dimensionality reduc- tion on IDS performance in IoT, highlighting PCA’s advantages in optimizing training time.