Optimizing Intrusion Detection in IoT through a Combination of Feature Selection and Deep Feedforward Neural Network

Abstract

Intrusion detection is a crucial aspect of maintaining the security of computer network devices. With the advancement of Internet of Things (IoT) technology, the need for intrusion detection has become even more critical due to the many IoT devices that remain inadequately protected due to resource constraints. Although various intrusion detection methods have been developed, many of them have not been optimized to address the resource limitations of IoT devices, such as limited computational capacity and low power consumption. Furthermore, existing feature selection methods often overlook the potential of combining various selection techniques to enhance accuracy and efficiency. This research introduces a novel intrusion detection framework that integrates multiple feature selection algorithms—Pearson Correlation, Spearman Correlation, and Mutual Information—during the pre-processing phase, aiming to reduce data complexity and improve classification accuracy. Unlike prior approaches that apply single-method feature selection, our method combines these techniques to capture both linear and non-linear dependencies, ensuring that only the most relevant features are retained. When evaluated on the UNSW-NB15 dataset, the proposed Deep Feedforward Neural Network (DFNN) classifier achieved an accuracy of 89.23%, outperforming other models in terms of accuracy and training efficiency. These improvements make the model better suited for real-world IoT environments with computational limitations.