ML Driven Client-Side Defense Mechanisms for Detection of Phishing based Web Spoofing Attacks

Statistics show that web spoofing attacks constitute nearly 56% of all cyber threats, with phishing attacks compromising around 30% of global data breaches. Current detection methods often rely on blocklists or heuristic rules, which lack adaptability and are less effective against rapidly evolving spoofing techniques. So, this paper introduces a novel client-side defence system called Client-Side Defense against Web Spoofing Attacks (CSD-WSA), leveraging comprehensive Uniform Resource Location (URL) feature extraction to identify malicious URLs with high precision. The proposed method analyzes essential URL components such as domain, path, query, and fragment, extracting 16 distinct metrics per component-including the frequency of dots, hyphens, slashes, and other potentially suspicious characters. Each extracted feature is preprocessed for optimal model performance, involving null value handling and scaling. Using Extreme Gradient Boosting (EGB), CSD-WSA achieves high classification accuracy in detecting phishing attempts, effectively distinguishing between legitimate and phishing URLs. Testing on a labelled URL dataset resulted in an accuracy of 99.14%, a precision of 99.15%, a recall of 99.14%, and an F1- score of 99.14%, underscoring the robustness of this client-side approach.