Key points are not available for this paper at this time.
Recently, we have noticed tremendous growth in the field of Information Technology. This increased growth has proliferated the use of new technologies and continued advancement of networking systems. These systems are widely adopted for real-time online and offline tasks. Due to this growth in information technology, maintaining security has gained huge attention as these systems are vulnerable to various attacks. In this context, an Intrusion Detection System (IDS) plays an important role in ensuring security by detecting and preventing suspicious activities within the network. However, as technology is overgrowing, malicious activities are also increasing. Moreover, legacy IDS methods cannot handle new threats, such as traditional signature-based methods requiring a predefined rule set to detect malicious activity. Also, several new methods have been proposed earlier to address security-related issues; however, the performance of these methods is limited due to poor attack detection accuracy and increased false positive rates. In this work, we propose and compare different deep-learning (DL) models that can be used to construct IDSs to provide network security. Details on convolutional neural networks (CNNs), Multilayer Perceptron (MLP), and long short-term memories (LSTMs) are introduced. A discussion of the outcomes achieved follows an assessment of the proposed DL model known as the FOA-CNN-LSTM technique. Comparisons are made between the suggested models and other machine-learning methods. This work presents a deep-learning approach based on hybrid CNN-LSTM with Fruit fly Optimization Algorithm (FOA) by ensemble techniques to distinguish between normal and abnormal behaviors.
Das et al. (Fri,) studied this question.