Abstract

The Internet of Things (IoT) connects devices to digital platforms to make our daily life better. Nonetheless with the expansion of IoT connectivity the dangers of network weaknesses and cyber threats also increase. Many notable Intrusion Detection Systems (IDSs) utilizing machine learning (ML) methods have been developed to handle this issue. Considering the resource limitation, in fog computing settings a lightweight IDS is crucial. This article presents a deep learning (DL) approach that integrates convolutional neural networks (CNN) and long short-term memory (LSTM) to develop an energy-efficient anomaly-based IDS. We made this model using a dataset targeting to minimize overhead while giving high accuracy and a low false rate. We assess the CICIoT2023, KDD-99 and NSL-KDD datasets to measure the effectiveness of the proposed IDS model using indicators such as latency, energy consumption, false alarm rate and detection rate. Our results show an accuracy exceeding 92% and a false alarm rate under 0.38%. These outcomes confirm that our system delivers security while maintaining efficient resource usage. The feasibility of implementing IDS with resources is shown through the successful operation of IDS features on a Raspberry Pi serving as a Fog node. The recommended lightweight model, taking up to 6.12 W of power, showcases its potential to function efficiently on devices with limited energy, such as low-power fog nodes or edge devices. We highlight energy conservation while giving accuracy, setting our method apart from current techniques. Comprehensive tests reveal a decrease in false positives, guaranteeing precise detection of real security threats and reducing superfluous notifications.