CyberShield: An Intelligent System for Threat Detection and Secure Message Recover
Abstract
Email-based attacks, such as phishing, spam, email-borne malware and malicious PDF attachments are becoming increasingly fast and sophisticated, and this requires a multi-layer approach to security. In this study, CyberShield, a scalable and multimodal threat identification system for a comprehensive examination of Gmail messages was introduced, which incorporates the content of the text, embedded URLs, metadata, sender behavior and pdf attachments. OAuth 2.0 is used to authenticate users in the system, and emails are retrieved to extract a variety of details, such as subjects, timestamps, domains, urls and attachments, using the Gmail API. A preprocessing worldlow condenses the text of emails using NLP techniques to look for domain related irregularities, and to scan PDF layouts to detect hidden scripts or suspicious content. All PDFs received through mailbox and all user uploaded PDFs to CyberSchield too are added, to detect both internal and external threats. A key component of the system would be Hybrid Ensemble Classification Module which will be an amalgamation of machine learning algorithms (Naive Bayes, Logistic Regression, SVM, KNN, Random Forest, XGBoost) and deep learning models (LSTM, CNN, RCNN) for linguistic, structural, and visual threat signature detection. The other form is the stacking-oriented meta-classifier which generalizes the results by giving a threat probability measure that increases the immunity against the newly emerging threats. The system classifies emails as safe, heuristics, recoverable in whitelist/blacklist and in user feedback loop to reduce false positives. The performance of CyberSchield, a hybrid ensemble, is reported to be 98.4% accuracy, 97.2% precision, 99.2% recall, and 98.2% Fl-score based on the results of an experiment based on a benchmark email dataset. A supplement to ROC curves, visualization also helps in visualizing the discriminative power of the proposed framework, that is, by visualizing the ROC curves and a coloured heatmap-like confusion matrix. Overall, CyberSchield is a flexible and comprehensive email security environment which can detect sophisticated threats in real time.