An Automated System To Detect Social Engineering Attacks Using ML Algorithm

Abstract

Socio technical threats remain a major problem in cybersecurity, especially since they involve getting information relating to security or even making the people perform a security-oriented act. This thesis proposes a new deep learning method, named SEAP (Social Engineering Attack Prevention), for discovering and protecting against such attacks. As the nature of the SEAP model is based on the unsupervised pre-training and supervised fine-tuning, it results in the above-mentioned benefits of higher capacity of the system to identify nonlinear, latent relation ships connected with the occurrence of social engineering attacks. Concerning the SEAP archi tecture, we also have ConvMix blocks to improve the detection while not making calculations of large datasets unmanageable. The employed dataset in this study is Phishing which comprises of record 10000 and predictor 50; which includes URL syntactical structure and content. Data preprocessing is then carried out on the database to prepare it for training involving process including scaling or feature engineering or normalization. In this respect, the Social Engineer ing Attack Prevention (SEAP) model pre-infuses a Deep Belief Network (DBN) architecture of multiple layers of Restricted Boltzmann Machines (RBMs). The self-adjusting phase of the variables and then learning of the parameters through the use of the Contrastive Divergence algorithm is then followed by an improvement in the classification through the application of the back-propagation supervised fine-tuning. The effectiveness of the SEAP model is checked through experiments and it meets high results, namely 96% of the accuracy with the help of the x dataset of the detection of phishing. Indicators of performance such as the exactness, the rate of recall, and the two types of the f-scores where each of them was equal to 0.96. Thus, while adopting the identification of anomalous behavior to the given subject, this thesis highlights the necessity of employing additional ML approaches to prevent social engineering attacks. Overall, there are two main advantages found in the novel SEAP model: this sugges-tion in its architecture and methodology is quite effective for a broad scale of enhancement of cybersecurity and can be employed efficiently in real-life scenarios like designing the system for detecting the phishing. Further investigative efforts can develop more of such approaches that improve the identification processes constituting the formation of secure and resilient cyber realms