Figure 1. ”Cyber security” and ”Deep learning” popularity
scores in a range of 0 (min) to 100 (max) through time, with the x-axis
representing timestamp information and the y-axis representing the
associated popularity score.
The popularity trend in Figure 1 is based on Google Trends data
collected over the previous five years. In this research, we consider
several standard neural networks and deep learning approaches in the
context of cybersecurity, including supervised, semi-supervised,
unsupervised, and reinforcement learning. These include I multilayer
perceptron (MLP), convolutional neural network (CNN or ConvNet), (ii)
recurrent neural network (RNN) or long short-term memory (LSTM), (iv)
self-organizing map (SOM), (v) auto-encoder (AE), (vi) restricted
Boltzmann machine (RBM), (vii) deep belief networks (DBN), (viii)
generative adversarial network (GAN) (DRL or deep RL). These deep neural
network learning techniques and their ensembles and hybrid approaches
can intelligently solve various cybersecurity problems, including
intrusion detection, malware or botnet identification, phishing, and
predicting cyber-attacks such as DoS fraud detection and cyber-anomalies
[7]. Construct security models because they are more accurate,
especially when learning from massive security datasets. This paper’s
contribution may be summarised as follows: