Abstract
An attempt by cyber fraud to use one or more machines across one or many networks is known as a cyber-attack. Not every website that a person visits can be completely safe from cyberattacks. Thus, the user must assure cyber-security. By creating two distinct Deep Learning (DL) models that can recognize the existence of a cyberattack on a website, this study seeks to aid the user. For this purpose, the NSL-KDD dataset is used to build a database with more than 40 features related to the detection of a cyberattack. For optimal performance, the database has been preprocessed. Normalization, data encoding, and statistical highlighting are all parts of the preprocessing. The preprocessed dataset is then divided into its training and testing halves. Two different deep-learning algorithms generated two distinct models. Convolutional Neural Network (CNN) and Hybrid Attention Neural Network (HANN) are the algorithms employed in this investigation. The most effective DL model for detecting a cyberattack is identified once it has been trained and tested. Both models displayed various performance categories while being trained. The accuracy values of the model created using the CNN algorithm increased virtually linearly, reaching a point with an accuracy of more than 90% after the 14th epoch. Yet, during the 6th epoch alone, the model created using the HANN algorithm achieves an accuracy higher than 90%. During the first training period, the HANN model's highest loss value is just under 90%. The final accuracy of the model created using the CNN algorithm is 90.8%, whereas the accuracy of the HANN model is 95.8%, according to testing. In the end, it may be said that the HANN algorithm is more effective at spotting cyber-attack than CNN. This concept may one day be implemented as a backend processor for a website that verifies the legitimacy of other websites and software.
Keyword
Cyber-Attacks, Cyber-Security, Data Encoding, Filtering, Normalization, Deep Learning Models.
PDF Download (click here)
|