A review of various challenges in cybersecurity using Artificial Intelligence
Harsh Chaudhary, Ankit Detroja, Priteshkumar Prajapati, Parth Shah
Abstract
Cybersecurity using Artificial Intelligence is a double-edged sword, it can improve security substantially but it also creates a possibility of new forms of attack, which performed on Artificial Intelligence itself. Machine Learning algorithms are proved useful at identifying zero-day attacks or detecting an unusual behavior of systems that might indicate an attack or a malware. This research work has reviewed various security threats and defensive techniques, open challenges in cybersecurity domain for intrusion detection, malware detection and network anomaly detection systems using various Machine Learning and Deep Learning algorithms. It is found that most of the discussed approaches used supervised models. For intrusion detection, RBF-SVM (Radial Basis Function - Support Vector Machine) model gave highest accuracy of 99.90% while in malware detection DNN (Deep Neural Network) model gave 97.79% accuracy. For pirated software identification, a DNN model was used and it gave 96% accuracy. Seq2Seq (Sequence-to-Sequence) model worked best for network anomaly detection giving an accuracy of 99.90%. On the other hand, for anomaly detection a DBN (Deep Belief Networks) based model is used which gives 69.77% accuracy. Finally, this paper discusses about the 5G's security, cyber-attacks and the major role of the above emerging fields in the future of cybersecurity.