A Genetic Algorithm based Feature Selection and CNN based Ensemble Model for Intrusion Detection in IoT Smart Environments

Abstract

The growing number of Internet of Things (IoT) devices has resulted in a significant surge in network attacks, frequently causing harmful and catastrophic consequences. Malicious actors may utilize these devices to infiltrate the network infrastructure by taking advantage of hardware and software weaknesses through uninterrupted internet access. Despite significant advancements in the field of network IDS (Intrusion Detection System), there is still a lack of employing intrusion detectors in IoT environments. Hence, to address this issue, a neural network model-based intrusion detection system is introduced, which can effectively detect and classify various types of attacks on IoT devices used in intelligent applications. A feature reduction technique and a hyperparameter optimization strategy to reduce both the computing time and overhead were utilized. Important features chosen via a genetic algorithm-based feature selection model are transformed into colour images for use as input to several Convolutional Neural Network (CNN) architectures, including Xception, VGG16, and VGG19 models. The suggested ensemble model, which combines Xception, VGG16, and VGG19 classifiers using a genetic algorithm to select the most relevant features, is 98.7% accurate, which is 5% better than individual classifiers. This novel approach significantly reduces false positives while cutting computational latency when compared to existing models. By optimizing both detection speed and accuracy, the proposed system enables real-time intrusion detection, offering a scalable and efficient solution for securing IoT devices in smart environments. These advancements underscore the system’s potential to set a new standard in IoT security.