Article Preview
Top1. Introduction
The design of cloud computing is for allowing the on-demand, appropriate, universal network contact to a common configurable computing property that can quickly release and provisioned with service provider interaction or minimal management effort (Nagar, U., (2017), Nagarajan, N and Thirunavukarasu, R. (2020)). Security is an important concern in cloud computing for the recent applications of virtual machine monitor (VMM) because of several susceptibilities (Jin, H., (2013)). The cloud computing environment is under threat from various types of cyber-attacks. Malicious users can inject malicious code into user's websites and gain unauthorized access to user's confidential information stored in the cloud (Dhote, P.M., (2018)), and hackers are continuing to develop new techniques to exploit and steal user credentials without the user's knowledge (Dhote, H and Bhavsar, M.D. (2018, April)). Nowadays, different IDS software and methods are developed specifically for the cloud environment to discover attacks. An efficient IDS method should detect various challenging attacks such as Probe, R2L, U2R, and Denial-of-Service (DoS). But R2L attack is a more challenging one due to its relations with both host and network-level features. Moreover, the U2R attack involves semantic information and therefore the detection is difficult near the beginning (Le, T-T-H., (2019)). In addition, the conventional IDS reduced the recognition accuracy and increased the false alarm rate with weak adaptableness. Hence, there is a requirement for an alternative and efficient IDS system. Detecting many attacks and reducing the false alarm rate is the recent popular research topic on IDS (Ouffoué, G., (2016, June), Bharati, M and Tamane, S. (2017, October)).
Nowadays, Decision tree, random forest, Support vector machine (SVM), Naïve bayes system and other machine learning approaches such as Artificial neural network (ANN) have been utilized in the domain of intrusion detection systems (Nathiya, T., and G. Suseendran. (2018), Negandhi, P., (2019)). But every approach holds certain own benefits and limitations. They do not provide good performance on all types of attacks. Hence, IDS has been changed by taking the advantages of deep learning approaches because these approaches use different transforms to excerpt high-dimensional hierarchical structured features (Ahmed, H.I., (2021)). Some of the existing deep learning-based intrusion detection approaches are Deep neural network, Recurrent Neural Network (RNN) and Deep Autoencoder (DAE) (Yin, C., (2017), Wang, W., (2017)). The existing machine learning approaches are not performing well because of the absence of representative feature extraction ability and the necessity of additional timing to train. Alternatively, deep learning methods derive high-dimensional features and it leads to dimensionality issues. To tackle this issue, the input patterns should reduce their dimensionality. Similarly, the present deep learning-based intrusion detection systems focused only on global accuracy, however they failed to detect intrusion while using small-scale data. Thus, a smaller proportion of data is considered in this research to prove the detection efficiency.