Developing a Hybrid Algorithm that Combine Rule-based and Machine Learning Methods to Detect Social Engineering Attacks

Social engineering is a science that utilizes social interaction to persuade an individual or organization to comply with an attacker’s request via a computer-related entity. Recently, social engineering attacks have grown rapidly, exploiting human naivety to gain unauthorized access to systems and information. Many models have been developed to address this problem. The hybrid algorithm proposed in this study is based on automating most cases of the optimized model for manual detection of social engineering attacks, while applying a dynamic blacklist algorithm to verify each line of text generated by a potential attacker, matching with dynamic topic blacklist, and employing natural language processing and machine learning, the proposed hybrid system successfully automated the detection of social engineering attacks. This achieved substantial operational efficiency, demonstrating the efficacy of the proposed algorithm as a defensive tool. The dataset used consists of 50,000 emails, collected from the public repository (Kaggle), balanced samples were integrated into a unified test file to feed the proposed hybrid system and measure its security response. The accuracy reached 0.9740 despite ignoring other features, which is considered excellent.